AI visibility analysis:can superintelligence readyour site?
We read your site from the outside with the identities of superintelligence agents and give you the results of 22 checks, in order of what to fix, on a single page.
For the agents of ChatGPT, Claude, Gemini and Perplexity
The analysis did not finish this time:
Email is optional. We use your address only to get back to you about this site.
Measurements
We do not ask you to sign up or give an email. You can share the report link with anyone you like; the report is closed to search engines.
Your AI search visibility test report opens like this.
- Access and indexing42 / 47
- Structured data6 / 12
- Page signals9.5 / 11
- Citation-ready content15.5 / 30
Readiness by engine
- ChatGPT68%
- Google AI Overviews · AI Mode68%
- Perplexity63%
- Claude65%
Fix first
- Content readable without JavaScript+5
- The opening paragraph answers directly+5
- Author and expertise visible+5
Target after these three fixes: 88 · Cited
What the report covers
- Score, level and a view of your siteA score out of 100 and one of five levels, next to desktop and phone views of your home page.
- Readiness for four enginesSeparate percentages for ChatGPT, Google, Perplexity and Claude, naming what is missing.
- How AI describes youAn AI model reads the text of your home page and summarizes your company, including what it could not work out.
- Fix order and target scoreThe points each fix is worth, where it goes on your site, with example code you can copy.
- 11 AI agents and 3 search crawlers at the doorWhich AI agents can get in, from your robots.txt rules and a live request.
- All 22 checks with evidenceEach result shows the value we read and the target; download the report as a PDF or send it to your inbox.
The figures in this sample are illustrative and do not belong to a real site. The score, area totals and engine percentages were calculated with our analysis formula from a sample set of results; your own report is filled with what we read from your site.
Is your site in ChatGPT? We check each engine separately.
-
ChatGPTChatGPT search and citations
- At the door
- The robots.txt rule for
OAI-SearchBotand a live request to your home page with the ChatGPT identity. - Weighs most
- OAI-SearchBot access14%
- Page opens for agents14%
- Text readable without JavaScript14%
- Fact density11%
- 12 checks across the areas
-
GoogleAI Overviews · AI Mode · Gemini
- At the door
- AI Overviews and AI Mode draw on the Google Search index: the door is the
Googlebotrule and no noindex or nosnippet on the page.Google-Extendedis for the Gemini app; it does not affect AI Overviews. - Weighs most
- Organization schema10%
- Service / Product schema10%
- Title and description10%
- Page opens for agents6%
- 17 checks across the areas
-
PerplexityPerplexity answers
- At the door
- The robots.txt rule for
PerplexityBot. - Weighs most
- Text readable without JavaScript17%
- Fact density17%
- Page opens for agents13%
- Last-modified date13%
- 9 checks across the areas
-
ClaudeClaude web search
- At the door
- The robots.txt rule for
Claude-SearchBotand a live request to your home page with that identity.ClaudeBotis a training crawler and does not affect the score. - Weighs most
- Text readable without JavaScript20%
- Page opens for agents15%
- Fact density15%
- Answer-first opening15%
- 9 checks across the areas
For each engine the report gives a readiness percentage: how much of what that engine weighs is in place. It is not a ranking prediction; if your robots.txt blocks the engine's agent, the percentage is capped at 25. Measuring how often answers mention and cite you is the job of a deep analysis.
robots.txt and llms.txt:
AI bot access check.
Visuals and slogans influence visitors; a superintelligence agent looks at your page’s raw text, headings and structured data. That is what we look at too, in the same order.
- 01HomeWe open it with the identity of ChatGPT’s search bot (OAI-SearchBot); if the door is closed, once more as an ordinary browser.
- 02robots.txtWe read your door decision for 11 AI agents and 3 search engine crawlers.
- 03llms.txtWe look for the summary you left for AI.
- 04SitemapWe try the one declared in robots.txt, or common names if there is none.
- 05With the Claude identityWe check whether Claude’s search bot (Claude-SearchBot) can get in.
- 06ScoreWe score access, indexing, content and structured data in four areas.
We change nothing on your site; we only read public addresses. We do not ask answer engines about you; we use a language model (OpenAI) for the summary and the suggestions.
Your AI SEO audit score comes from four areas.
Measurement setup (GA4, ad pixel) is not part of the score; the report shows it in a separate "Measurement readiness" box.
01 · 6 checks · 47 points
Access and indexing
Can AI agents get through the door, read your page without running JavaScript, and can Google show your page from its index?
ChatGPT search bot (OAI-SearchBot) can access the siteOAI-SearchBot rule in robots.txt8 points
- Target
- allowed
- Our recommendation
- Do not block OAI-SearchBot in robots.txt. It is the bot ChatGPT uses for search and for citing sources; it is separate from the GPTBot training crawler.
Google-Extended not blockedGoogle-Extended rule in robots.txt2 points
- Target
- not blocked
- Our recommendation
- Google-Extended controls whether your content can be used in Gemini models and for grounding answers. It does not affect Google Search or AI Overviews, so blocking it only restricts the Gemini side.
Page opens for AI agents (200)the home page is opened with ChatGPT and Claude identities10 points
- Target
- 200 for both agent identities
- Our recommendation
- Review any blanket “AI bot” block in your firewall or bot management; if legitimate agents are blocked, you cannot appear in answers at all.
Content readable without JavaScriptword count in HTML without JavaScript10 points
- Target
- 800 words on average · minimum target 300 (partial: 80 and above)
- Our recommendation
- Render the main text on the server. Most superintelligence agents do not run JavaScript; a page that looks empty will not be cited.
Sitemap publishedsitemap declared in robots.txt or found under a common name5 points
- Target
- published
- Our recommendation
- Publish an XML sitemap that covers all service and content pages, and declare it in robots.txt.
Can be shown in Google Search and AI Overviewsmeta robots, X-Robots-Tag and the Googlebot rule in robots.txt12 points
- Target
- no noindex, nosnippet or Googlebot block
- Our recommendation
- Remove the noindex or nosnippet directive on the home page (meta robots, meta googlebot, X-Robots-Tag) and the Googlebot block in robots.txt. Google cannot cite a blocked page in Search or in AI Overviews.
02 · 6 checks · 12 points
Structured data
Do you state who you are and what you offer with structured data? It helps Google rich results and Bing; no direct effect on AI citations has been shown, so its share is small.
Organization markupOrganization node in JSON-LD4 points
- Target
- present
- Our recommendation
- Add Organization JSON-LD to the home page with your name, address, logo, contact details and social profiles; Google and Bing use it to match your brand.
Social and company profiles linked (sameAs)sameAs links in Organization3 points
- Target
- at least 2 profiles
- Our recommendation
- Add verifiable profiles such as LinkedIn and Wikipedia/Wikidata to the Organization schema as sameAs; mentions of your brand on other sites can then be matched to your site.
Service/Product markupService or Product markup2 points
- Target
- present
- Our recommendation
- Add Service (or Product) JSON-LD to every service or product page with name, description, provider and service area.
FAQ markup (FAQPage)FAQPage markup1 points
- Target
- present
- Our recommendation
- Write an FAQ section that answers real customer questions; FAQPage markup describes the section to Bing and Google. Having the answer written plainly on the page matters more than the markup.
Breadcrumb markup (BreadcrumbList)BreadcrumbList markup1 points
- Target
- present
- Our recommendation
- Declare the page’s place in the site with BreadcrumbList.
Structured data in JSON-LDJSON-LD or microdata1 points
- Target
- JSON-LD (partial: microdata only)
- Our recommendation
- Publish structured data as JSON-LD instead of microdata; it is the easiest format to read.
03 · 5 checks · 11 points
Page signals
Is your page’s identity (title, description, single H1, canonical, language) clear and unambiguous?
Title and description tags written<title> and meta description4 points
- Target
- title and description (partial: title only)
- Our recommendation
- Write a unique <title> and a 120–160 character meta description for every page.
One meaningful H1number of H1s on the page2 points
- Target
- one H1
- Our recommendation
- Use a single H1 that says what the page is about.
Canonical points to the right URLcanonical URL3 points
- Target
- canonical pointing to the same domain
- Our recommendation
- Add a self-referencing canonical tag to every page; duplicate URLs split visibility.
Sharing tags (Open Graph) setog:title, og:description, og:image1 points
- Target
- 3/3
- Our recommendation
- Add og:title, og:description and og:image tags; social previews use them, and some agents may read them when summarizing the page.
Page language and language versions declaredhtml lang and hreflang1 points
- Target
- lang declared
- Our recommendation
- Set <html lang="…">; if there are versions in other languages, link them with hreflang.
04 · 5 checks · 30 points
Citation-ready content
Is your content ready to be quoted: an opening paragraph that answers up front, facts, a recent date and an author.
The opening paragraph answers directlywhere the opening paragraph sits and whether it says what you do10 points
- Target
- an opening paragraph at the top of the page that states what you do (partial: one of the two)
- Our recommendation
- Open the page with a paragraph that says who you are and what you offer in one or two sentences ("X is a … that provides …"). Most quotes in AI answers come from the first part of a page.
Fact-dense content (figures, measures, durations)measures and numbers in the text7 points
- Target
- 8 or more measures/numbers (partial: 3–7)
- Our recommendation
- Write facts instead of slogans: durations, price ranges, scope, measures, numbers. Text with facts is associated with being cited in answers; a slogan carries no information to quote.
Content is current (dated within the last year)dateModified, article:modified_time, lastmod and the age of the date7 points
- Target
- a date within the last year (partial: older date)
- Our recommendation
- Declare publish and update dates on your pages (dateModified, lastmod in the sitemap) and keep content current; AI answers tend to pick newer sources.
Author and expertise visibleauthor or person markup5 points
- Target
- author markup
- Our recommendation
- Show who wrote the content (author name, Person schema, author page); expertise signals increase trust.
/llms.txt published/llms.txt file1 points
- Target
- published, longer than 50 characters
- Our recommendation
- If you like, publish an /llms.txt with a short company summary and links to key pages. It does no harm, but no effect on visibility has been shown; Google does not use it, some coding tools read it.
Five levels · from Invisible to Cited
Invisible: AI agents most likely cannot read your site, or your page does not appear in the index.
This section is a sample GEO strategy study for an anonymised client; it is not part of your report.
Sample GEO strategy study · anonymised client
GEO audit example: we asked four engines 19 questions for a diamond jewelry brand.
- 76answers · 19 questions × 4 engines, October 9, 2026
- 8answers mentioned the brand (10.5%)
- 12answers cited the brand's site (15.8%)
- 11/19questions where it appeared in at least one engine
| Question | ChatGPT | Gemini | Perplexity | Claude | Competitors named |
|---|---|---|---|---|---|
| What should I look for when choosing a solitaire diamond ring for a proposal? | ChatGPT: not yet | Gemini: cited | Perplexity: not yet | Claude: not yet | — |
| How much do solitaire diamond rings cost in 2026? | ChatGPT: mentioned | Gemini: cited | Perplexity: not yet | Claude: not yet | Competitor F, Competitor A +2 |
| What determines the price of a 1-carat solitaire diamond ring? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: not yet | — |
| Oval or round solitaire: which looks bigger on the finger? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: not yet | — |
| Is a baguette ring or a solitaire better for an engagement ring? | ChatGPT: not yet | Gemini: cited | Perplexity: not yet | Claude: not yet | — |
| What is the difference between a five-stone and a full eternity diamond ring? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: not yet | — |
| What are the 4Cs of diamonds, and how do color and clarity affect price? | ChatGPT: not yet | Gemini: cited | Perplexity: not yet | Claude: not yet | — |
| Why does a certificate matter when buying a diamond, and which one should I look for? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: not yet | — |
| How can you tell a real diamond? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: not yet | — |
| What should I look for when buying a diamond necklace as a gift? | ChatGPT: not yet | Gemini: cited | Perplexity: not yet | Claude: not yet | — |
| How much does a diamond tennis bracelet cost? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: not yet | Competitor D, Competitor A +1 |
| How do I choose diamond earrings for my face shape? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: not yet | — |
| Is it safe to buy diamonds online, and what should I watch out for? | ChatGPT: not yet | Gemini: cited | Perplexity: not yet | Claude: not yet | — |
| Are there diamond brands where I can trade in my old ring for a bigger carat model by paying the difference? | ChatGPT: not yet | Gemini: mentioned and cited | Perplexity: mentioned and cited | Claude: mentioned and cited | Competitor B, Competitor C +2 |
| Which diamond shops in Istanbul's Nuruosmaniye can be trusted? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: mentioned and cited | Competitor A, Competitor B +1 |
| Which shops in Ankara can I trust to buy a diamond ring? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: mentioned and cited | Competitor A, Competitor B +4 |
| Where can I buy a solitaire diamond ring in Izmir? | ChatGPT: not yet | Gemini: not yet | Perplexity: not yet | Claude: not yet | Competitor D, Competitor F +3 |
| Which brands are alternatives to Competitor A for a solitaire diamond ring? | ChatGPT: not yet | Gemini: mentioned | Perplexity: not yet | Claude: mentioned | Competitor B, Competitor E +6 |
| What should I look for when buying a diamond necklace? | ChatGPT: not yet | Gemini: cited | Perplexity: not yet | Claude: not yet | — |
By engine · out of 19 answers
- ChatGPT1 mentions · 0 citations
- Gemini2 mentions · 8 citations
- Perplexity1 mentions · 1 citations
- Claude4 mentions · 3 citations
Most cited sites
- Another jewelry site 121
- Competitor A site19
- Competitor D site17
- Competitor F site15
- Competitor B site13
- Another jewelry site 213
- The brand's site12
- Competitor G site12
Count: answers that cited the site.
When a service has its own page, the engines find it
For the question about trading in an old ring by paying the difference, Gemini, Perplexity and Claude both named and cited the brand. The cited page was the one dedicated to that service. We are carrying the same pattern over to other services.
An existing guide is the closest opportunity to be cited
The site already has a guide answering "How can you tell a real diamond?". We are adding a short paragraph that answers the question directly at the top, plus current details, and checking the same question in the next measurement.
Measured on October 9, 2026 in app.markention.com through the engines' APIs (ChatGPT gpt-4.1-mini, Gemini 2.5 Flash, Perplexity Sonar, Claude Haiku 4.5); answers in the consumer apps can differ. We track the trend by repeating the same question set every month. Brand and competitor names are hidden.
After the AI analysis, the next step is yours.
-
01
Apply the report yourself
Fixes are ordered by the points they are worth; most of them happen in robots.txt, schema and page copy. Download the report as a PDF or share its link with your team.
-
02
Let us read it together
In a 45-minute Google Meet call with Cihat Murat Yüksel, who leads our operation, and our team, we can go through the results on your site and set priorities for your brand together.
-
03
Deep analysis and GEO work
We measure how often ChatGPT, Gemini, Perplexity and Claude mention and cite you for the questions your audience asks, side by side with your competitors, and review your service and product pages one by one.
AI visibility guide: how brands get recommended.
Contents 12 sections
- 01What is AI visibility and why measure it?
- 02How does AI find and recommend brands?
- Training data as the model’s memory
- Live search and the cited part of the answer
- Which pages get cited
- Entity and consistency
- 03How GEO differs from SEO and where they overlap
- 04How ChatGPT, Gemini, AI Overviews, Perplexity, Copilot and Claude differ
- 05How is AI visibility measured?
- 1. A question set based on what your audience asks AI
- 2. Mention rate, citation rate and share of voice
- 3. Google AI Overviews and AI traffic in Analytics
- 06What does an AI visibility analysis include?
- 07Technical requirements for AI to read your site
- robots.txt and AI crawlers
- JavaScript-dependent content
- Structured data
- Speed, indexing and preview controls
- llms.txt is not an official standard
- 08Content and reputation on third-party sites, reviews and Wikipedia
- 09Common mistakes in AI visibility
- 10AI visibility checklist
- 11Our process from free analysis to measured visibility
- 12Scoring handbook: how the 100 points are allocated
- Method: what we measure and what we do not
- The four areas and why they are weighted this way
- A · Access and indexing: 47 points
- B · Structured data: 12 points
- C · Page signals: 11 points
- D · Citation-ready content: 30 points
- What do the score levels mean?
- Version history: from version 1 to version 2
- Limits: what this score does not tell you
- Sources behind the scoring
The AI visibility analysis reads your site as AI agents do and scores 22 checks in four areas: access and indexing (47 points), structured data (12), page signals (11) and citation-ready content (30). We explain the reasoning behind each weight in the scoring handbook at the end of this page. It reads the home page, robots.txt, llms.txt and the sitemap, plus up to 4 sample pages from the sitemap, takes a home page screenshot and measures speed with PageSpeed (speed is not scored). It does not measure whether your brand is mentioned in AI answers; that is the subject of the deep analysis.
AI visibility means your brand is mentioned, recommended and cited in the answers of such as ChatGPT, Gemini, Google AI Overviews, Perplexity, Copilot and Claude. AI now acts as an adviser beside the search box; when someone asks “which company should I work with?”, the names in the answer matter as much as ranking.
Drawing on official documentation, this guide explains how AI systems find a brand, why they cite a page, how visibility is measured and what to change on your site. The AI visibility analysis above is the first step: it reads your site as AI agents do and lists the gaps in fix order.
The first condition for appearing in AI answers is a site that is open to search bots and indexed; no special file or hidden markup replaces this.
What is AI visibility and why measure it?
AI visibility is the sum of the questions, contexts and sources with which your brand appears in AI answers. Classic search measures position and clicks; answer engines involve three separate events, each measured independently:
- Mention: The brand name appears in the answer text (“for …, X and Y stand out”).
- Citation: A page on your site is linked below or within the answer.
- Context: The brand appears as a recommendation, in a comparison or with a warning.
A brand can be mentioned without being cited, and cited without being named. So “does ChatGPT recommend my brand?” cannot be settled with one try. People get different answers to the same question, so we measure against the questions your audience actually asks, not a single query.
The reason to measure is simple: answer engines often give the user one recommendation. Tenth place on a results page still gets the odd click; absence from an AI answer means the brand does not exist for that question.
How does AI find and recommend brands?
An answer engine knows your brand from two places: what the model learned in training and what it reads on the web at question time. The two work differently, change at different speeds and differ in how much you can influence them.
Training data as the model’s memory
Large language models build general world knowledge from the texts they read in training. Much of this text comes from the open web; public archives such as Common Crawl are among its known sources [15]. A model cannot remember a page published after its training. If your brand is rare or inconsistent in that data, the model may not recognize you, may repeat outdated facts or may confuse you with a competitor. Training data cannot change quickly; the work here is keeping the brand’s web footprint consistent and accurate over the long term.
Live search and the cited part of the answer
For current, commercial and local questions, answer engines search the web at question time, read what they find and build the answer from it. This is called . Google writes that AI Overviews and AI Mode run several related searches across subtopics (query fan-out) to build an answer [1]. For Gemini, Google calls this “grounding”: giving the model content from the Google Search index at question time [3].
This is the side your brand can influence quickly. To be a source in live search, a page must be open to that engine’s crawler, indexed and clear in answering the question. Google states it plainly: a supporting link in an AI Overview only needs to be indexed and snippet-eligible; there are no extra requirements or special optimization [1].
Which pages get cited
No answer engine publishes its source selection algorithm in detail, but published documents and our measurements agree. Cited pages share four traits:
- A section that answers the questionThe heading carries the question; the first sentence below gives the answer. Bing also recommends putting core information at the top of the page without a long introduction [9].
- Verifiable factsNumbers, dates, conditions and evidence are stated clearly. In the GEO study, adding sources, quotations and statistics raised visibility by up to 40% on its test sets [13].
- One topic, one URLEach page covers one topic in depth; two pages answering the same question split their strength.
- A trustworthy brand footprintWho owns the page, when it was updated and how other sources describe the brand are read from outside the page.
Entity and consistency
AI systems recognize a brand not just by its domain but as an : its name, what it does, where it is and who runs it. Google writes that Knowledge Graph facts come from various sources that compile factual information, and that organizations can claim their knowledge panels and suggest changes [5]. If your brand name differs across your site, Google Business Profile, social accounts and industry directories, the model struggles to merge them into one entity. sameAs links in Organization schema tie these profiles to one machine-readable identity.
How GEO differs from SEO and where they overlap
GEO (generative engine optimization) is the work of getting a brand mentioned and cited in AI answers; see What is GEO for a full definition. Google’s own guide says optimizing for generative AI search is optimizing for the search experience, so it is still SEO [2]. We do not split the two across teams; they differ in goal and metrics.
| Topic | SEO | GEO |
|---|---|---|
| Goal | A top position in the results list | Being mentioned and cited in the answer |
| Unit | Page | Passage, sentence, fact |
| Metric | Impressions, clicks, average position | Mention rate, citation rate, share of voice, context |
| Competitor | Sites ranking for the same keyword | Brands mentioned and sites cited for the same question |
| Common ground | A crawlable, indexed site, content that answers the question clearly, a consistent brand across the web | |
In practice, all technical GEO work also helps SEO, but not always the reverse. A well-ranking page that opens with advertising copy may lack a quotable sentence. We handle both in one plan in our SEO consulting and GEO consulting work.
How ChatGPT, Gemini, AI Overviews, Perplexity, Copilot and Claude differ
Each answer engine reaches the web through its own crawler or a search index, so site owners’ controls and data vary. The table compiles what each company states in its own documentation.
| Surface | Live web access | robots.txt token | Data visible to the site owner |
|---|---|---|---|
| Google AI Overviews and AI Mode | Google Search index; multiple searches across subtopics | Googlebot (Search); snippet controls apply | Part of overall Web traffic in Search Console, no separate row [1] |
| Gemini apps | Grounding from the Google Search index | Google-Extended (training and grounding); does not affect Search [3] | gemini.google.com referral in Analytics |
| ChatGPT | Web search | OAI-SearchBot (search), GPTBot (training), ChatGPT-User (user request) [6] | chatgpt.com referral in Analytics |
| Perplexity | Own index and visits on user request | PerplexityBot (search), Perplexity-User (user request) [8] | perplexity.ai referral in Analytics |
| Microsoft Copilot | Bing index | Bingbot | Bing Webmaster Tools › AI Performance [10] |
| Claude | Web search | Claude-SearchBot (search), ClaudeBot (training), Claude-User (user request) [7] | claude.ai referral in Analytics |
The key distinction is that training and search bots use separate tokens. OpenAI writes that sites blocking OAI-SearchBot are not shown in ChatGPT search answers and that robots.txt changes can take about 24 hours to take effect, while blocking GPTBot asks that content not be used for training [6]. Anthropic says blocking Claude-SearchBot may reduce visibility in user searches [7]. Perplexity writes that PerplexityBot is not used to train foundation models and that Perplexity-User, which visits on user request, generally ignores robots.txt rules [8]. Citations in Copilot and Bing’s AI summaries appear in Bing’s AI Performance report, launched in February 2026, as total citations, cited pages and grounding queries [10]. For details, see Google AI Overview consulting and, for Copilot, Bing AI consulting.
How is AI visibility measured?
No answer engine gives site owners a full report of the questions they were mentioned in. We therefore combine three sources, building the measurement set with the method of Cihat Murat Yüksel, who leads our operations, after going through your whole site.
1. A question set based on what your audience asks AI
Our questions come not from automated templates but from your service and product pages, real Search Console queries, tracked keywords and questions about your brand. Comparison questions use only your real competitors, and our team reviews every question before measurement. A poor question set wastes budget and misleads.
2. Mention rate, citation rate and share of voice
Once every question has run on every engine, each answer is processed. Mention rate is the share of answers that include your brand. Citation rate is the share that cite your domain. Share of voice is the answers mentioning a brand divided by the mentions of all brands, putting you on one scale with competitors. Position in the answer and context (positive, neutral, negative) are recorded too. Echoes do not count: a brand name that appears only because the question contained it is not a mention.
3. Google AI Overviews and AI traffic in Analytics
Google does not report AI Overviews traffic separately in Search Console [1]. So our keyword tracking checks whether each tracked keyword’s results page shows an AI Overview and whether it cites your site. In Google Analytics 4, we split visits from chatgpt.com, gemini.google.com, perplexity.ai, copilot.microsoft.com, claude.ai and similar sources by engine and compare them with the previous period. We observe that some AI app visits carry no referrer and land in direct traffic, so the Analytics figure is a lower bound.
What does an AI visibility analysis include?
The analysis has two layers. The free analysis here measures your site’s technical readiness for AI agents; the in-depth analysis shows how you actually appear in answers and why competitors lead.
- Access as an agentWe open your home page as ChatGPT’s search bot and Claude’s crawler, and read what your robots.txt decides for nine AI agents.
- Text readable without JavaScriptWe count the text readable without running JavaScript; content the agent cannot see cannot be quoted.
- Structured data and identityWe check organization, service or product, FAQ and breadcrumb markup, plus title, description, a single H1, canonical and language tags.
- Citability signalsFacts and numbers in the text, author markup, last-updated date and the llms.txt file.
- Score and order of fixesThe results of 22 checks are scored out of 100 across four areas; the report starts with the fix worth the most points.
- In-depth analysisThe question set runs on four engines; mentions, citations and share of voice are compared with competitors, and for each gap we specify what to add to which page.
The free score rests on criteria we set, stated beside each result; it is not published by AI companies. It shows whether the site is ready and puts the fixes in order.
Technical requirements for AI to read your site
The technical side is the door to visibility: if it is closed, even excellent content stays out of answers. We check these areas:
robots.txt and AI crawlers
robots.txt is a protocol standardized by the IETF as RFC 9309; each crawler follows the rule group matching its name, and the rules are a request, not access authorization [11]. To be mentioned in answers, stay open to search bots (Googlebot, Bingbot, OAI-SearchBot, PerplexityBot, Claude-SearchBot). Allowing training bots (GPTBot, ClaudeBot, the Google-Extended control, CCBot) is a separate brand decision that does not directly decide search visibility. A frequent mistake is a security plugin or CDN rule blocking all bots. Our article What is robots.txt and our AI access control service help you decide bot by bot.
JavaScript-dependent content
Google processes JavaScript pages in its crawling, rendering and indexing stages [17]. We have seen no official commitment that other AI crawlers run JavaScript. So we recommend keeping core text such as the service description, pricing principles, contact details and FAQs in the initial HTML. Our analysis reads pages without JavaScript to show the gap.
Structured data
Google writes that AI features need no special schema.org markup and that structured data must match the visible text [1][4]. It is not a ranking trick but a machine-readable statement of who you are and what you offer. Organization (name, logo, address, sameAs), Service or Product, FAQPage and BreadcrumbList markup must match the page exactly.
Speed, indexing and preview controls
A slow or timed-out page is unreadable for an agent visiting on a user’s behalf too. Good Core Web Vitals thresholds are 2.5 seconds for LCP, 200 milliseconds for INP and 0.1 for CLS [16]. Speed is not scored but is shown at the top of the report for information. The nosnippet, data-nosnippet, max-snippet and noindex controls set how much Google’s AI features may show [1]; a stray noindex removes the page from every surface.
llms.txt is not an official standard
llms.txt is a short Markdown map a site offers to language models, an independent proposal published at llmstxt.org [12]; no standards body such as the IETF or W3C has approved it. Google writes that Search does not use these files and that creating one neither harms nor helps visibility in Google Search [2]. It is cheap and some tools read it, so we do not object; but llms.txt never replaces robots.txt rules, indexing or good content. If you have none, our analysis drafts one for your site.
Content and reputation on third-party sites, reviews and Wikipedia
Answer engines do not learn about a brand from its own site alone; which brand gets recommended often depends on how other sources describe it. The in-depth analysis lists each source that cites competitors but not you. The areas of work are:
- Industry directories and comparison articles: Sources often seen below answers are “best …” lists and comparison pages, where the brand should appear with accurate information.
- Reviews and business profiles: Name, address and service description on Google Business Profile and your industry’s known review platforms should match your site; unanswered negative reviews can turn the context negative.
- News and digital PR: Being cited for expertise in an independent publication strengthens the model’s link between the brand and a topic.
- Wikipedia and Wikidata: Wikipedia’s conflict of interest guideline discourages organizations from editing their own articles in their own interest and requires paid editing to be disclosed [14]. Changes are proposed on the talk page; pushing an article for a brand that does not meet the notability criteria damages its reputation.
Google’s guide gives a clear warning here too: chasing artificial “mentions” across the web is less useful than it seems [2]. We do not buy links or mentions; we aim for accurate descriptions where the brand is genuinely discussed. How language models work underlies all this; see What is an LLM.
Common mistakes in AI visibility
The mistakes we find most often can usually be fixed with a one-line setting or a single paragraph:
- Blocking the search bot instead of the training botBlocking OAI-SearchBot instead of GPTBot, or Claude-SearchBot instead of ClaudeBot, removes the brand from search answers.
- Leaving core text to JavaScriptIf the service description on screen is not in the HTML the agent reads, it cannot be quoted.
- A first paragraph that is only a slogan“Building the future together” answers no question; the first sentence should say what you do, for whom and where.
- Schema with invisible informationMarking up a review rating or price that is not on the page breaks structured data guidelines [4].
- Treating llms.txt as a cure-allThe file does not make up for a closed robots.txt or weak content.
- Deciding from a single tryBeing mentioned or not in one query shows no trend; it takes a set and time.
- Creating separate thin pages for AIGoogle writes that pages duplicated to influence answers may fall under its scaled content abuse policy [2].
AI visibility checklist
You can tick off the list below for your own site; your ticks are saved only in this browser. The analysis above checks most of the technical items for you.
Our process from free analysis to measured visibility
Answer engines have created a new search habit: people now ask some questions directly to AI, and debates about superintelligence (ASI: artificial superintelligence) suggest the habit will last. We build your brand’s place here step by step:
- Free AI visibility analysisEnter your site in the field at the top of the page and the report is ready in seconds; we ask for no sign-up or email address.
- MeetingWe go through the report together and discuss your priorities and real competitors.
- Question set and first measurementWe go through your site, build the question set and take the first measurement on four engines.
- Fixes and contentA work plan for technical access, structured data, page structure and brand footprint.
- Regular measurementThe same set is measured again at intervals, and questions gained and lost are reported.
For paid visibility inside ChatGPT, see ChatGPT Ads consulting; for the concepts of AI and superintelligence, see What is superintelligence.
Scoring handbook: how the 100 points are allocated
The free analysis scores 22 checks in four areas: access and indexing 47, structured data 12, page signals 11 and citation-ready content 30. The total is 100. We set the weights based on official documentation from Google, OpenAI, Perplexity and Microsoft and on independent citation studies. This section states plainly what each check measures, how many points it carries and why it carries them.
A score is only worth as much as your understanding of how it was calculated. Most readiness scores in our industry do not publish their weights; when a report says “72/100”, you cannot see what that 72 adds up to. We do the opposite: we publish every line of the analysis here, with its weight and its evidence. If a weight changes, this section changes the same day.
Method: what we measure and what we do not
The analysis reads your site from the outside, the way an AI agent would see it. It requests the home page, robots.txt, llms.txt and the sitemap, and it also loads the page using the ChatGPT and Claude agent user agents. It then runs 22 rules on the raw HTML it receives. It does not ask an AI model anything; the result is the output of the rules.
Each check has three possible outcomes. A pass earns the check’s full points, a partial result earns half, and a fail earns zero. The total is rounded to the nearest whole number. For example, a partial result on a 10 point check earns 5 points.
What the analysis does not measure matters as much as what it does:
- Real AI answers: This score does not tell you whether your brand is mentioned in ChatGPT, Gemini or Perplexity answers. It measures the technical and content prerequisites for being cited. Your actual place in answers is measured with a question set, which we do in the deep analysis.
- Mentions on other sites: Studies show that AI search leans heavily on independent third party pages, lists and press coverage. We cannot measure that external footprint by reading a single site, so it is not scored.
- Speed: Page speed appears at the top of the report for information but is not scored. We found no solid evidence that speed works as an AI citation factor.
- Rankings: We do not measure where you rank on Google. We measure whether you are indexed and eligible for snippets.
- Measurement setup: GA4 and ad pixels are not scored; they appear separately in the measurement readiness box below.
The four areas and why they are weighted this way
| Area | Points | Checks | Rationale in one sentence |
|---|---|---|---|
| A · Access and indexing | 47 | 6 | A page that cannot be read or is not indexed cannot be a source in any answer. |
| B · Structured data | 12 | 6 | Useful for Google and Bing; measurements have not shown a direct effect on AI citation. |
| C · Page signals | 11 | 5 | The title, description and canonical help search surfaces identify the page correctly. |
| D · Citation-ready content | 30 | 5 | Citations come from the top of the page and from factual, current text. |
Access and indexing (47). This is the heaviest area because everything else depends on it. Google writes that there are no additional requirements to appear as a supporting link in AI Overviews and AI Mode; a page only needs to be indexed and eligible to be shown with a snippet. OpenAI states that sites that block OAI-SearchBot will not be shown as sources in ChatGPT search answers. Most agents do not run JavaScript; if content is built only in the browser, the agent sees an empty page. In a study of Turkish search results, pages whose content was readable in the initial HTML were cited in Google AI Overviews 37.0% of the time, against 26.5% for pages that were not. In the same study, 75.3% of the sources in Turkish overviews came from the top 10 organic results. That is why access and indexing carry almost half of the total.
Structured data (12). Microsoft has said that schema markup helps its language models understand content. Google, on the other hand, writes that no special schema.org markup is needed for its AI features. In controlled comparisons, pages that added JSON-LD showed no meaningful difference in ChatGPT citations, and one independent study found the relationship between schema and citation to be statistically insignificant. So we cut the weight from 30 in version 1 to 12. We did not drop the area entirely, because structured data still does real work for Google rich results and on Bing.
Page signals (11). The title, description, H1, canonical and language attribute are the foundations of classic search. Because most AI answers rely on search indexes, these signals matter indirectly, but they do not decide citation on their own. The weight reflects that middle position.
Citation-ready content (30). A study of 1.2 million ChatGPT answers found that 44.2% of citations came from the first 30% of the page, and that definition sentences were twice as common in cited passages. Pages cited by AI assistants are on average 25.7% fresher than organic results. In a peer reviewed experiment, LLM rankers moved a passage up many positions when only its publication date was changed. The academic GEO study showed that text with added sources and statistics becomes more visible in answers, although the gain shrank under more realistic competitive conditions. These findings set the weight of this area.
A · Access and indexing: 47 points
| Check | Points | How to pass | Why it matters |
|---|---|---|---|
| ChatGPT search bot (OAI-SearchBot) can access the site | 8 | robots.txt does not block OAI-SearchBot. If robots.txt cannot be read, the result is partial. | According to OpenAI, a site that blocks this bot is not shown as a source in ChatGPT search. The training bot GPTBot is separate; blocking it does not affect this check. |
| Google-Extended not blocked | 2 | robots.txt does not block Google-Extended. | It affects only Gemini models and grounding in Gemini answers, not Google Search or AI Overviews. Hence the low weight. |
| Page opens for AI agents (200) | 10 | The home page returns 200 to the ChatGPT and Claude agent user agents. Both succeeding is a pass, one is partial. If a firewall stops our own request, we record a partial result, because we cannot verify from the outside whether the real bot gets through. | Even when robots.txt allows access, a blanket bot block in the firewall makes the site invisible. This is the silent failure we see most often. |
| Content readable without JavaScript | 10 | At least 300 words in the raw HTML is a pass; 80 to 299 words is partial. | Most agents do not run JavaScript. If the text is built in the browser, the agent sees an empty page. |
| Sitemap published | 5 | A readable XML sitemap exists, either declared in robots.txt or at a standard location. | Bots learn about your pages and their update dates from it. Its effect is indirect, so the weight is moderate. |
| Can be shown in Google Search and AI Overviews | 12 | The home page has no noindex or nosnippet (meta robots, meta googlebot, X-Robots-Tag), and robots.txt does not block Googlebot. | This is the only requirement Google states for AI Overviews. It is the highest single weight in the analysis. |
B · Structured data: 12 points
| Check | Points | How to pass | Why it matters |
|---|---|---|---|
| Organization markup | 4 | The home page has Organization (or a subtype) JSON-LD. | Google and Bing use it to match your brand with its name, logo and contact details. |
| Social and company profiles linked (sameAs) | 3 | At least 2 sameAs links is a pass; 1 link is partial. | Your brand’s profiles on other sites are recognized as the same entity as your site. |
| Service/Product markup | 2 | The home page has Service, Product, Offer, OfferCatalog or ProfessionalService markup. | It states what you sell in a machine readable form; it matters more on product sites. |
| FAQ markup (FAQPage) | 1 | FAQPage or QAPage markup is present. | What really matters is that the answer is written clearly on the page; the markup itself is a small addition. |
| Breadcrumb markup (BreadcrumbList) | 1 | BreadcrumbList markup is present. | It states where the page sits within the site. |
| Structured data in JSON-LD | 1 | JSON-LD is a pass; microdata only is partial. | It is the format Google recommends and the easiest to read. |
C · Page signals: 11 points
| Check | Points | How to pass | Why it matters |
|---|---|---|---|
| Title and description tags written | 4 | A title and a meta description together is a pass; a title alone is partial. | It is the first text shown on search surfaces and says in one line what the page is about. |
| One meaningful H1 | 2 | Exactly one H1 is a pass; more than one H1 is partial. | It gives a clear signal of the page’s main topic. |
| Canonical points to the right URL | 3 | A canonical pointing to your own domain is a pass; one pointing to another domain is partial. | Duplicate URLs split visibility, and a wrong canonical hands your page over to another address. |
| Sharing tags (Open Graph) set | 1 | og:title, og:description and og:image all present is a pass; one or two is partial. | Social previews use them; their effect on agents is weak. |
| Page language and language versions declared | 1 | The <html lang> attribute is present. If hreflang exists, it is shown in the report. | It makes it easier to match a question in one language with sources in the same language. |
D · Citation-ready content: 30 points
| Check | Points | How to pass | Why it matters |
|---|---|---|---|
| The opening paragraph answers directly | 10 | An opening paragraph of at least 15 words sits within the first 30% of the text and contains a definition sentence saying who you are or what you offer. Both is a pass; one is partial. | Most quotes in AI answers come from the top of the page, and definition sentences are quoted more often. |
| Fact-dense content (figures, measures, durations) | 7 | At least 8 figures or measures on the page (durations, prices, percentages, counts and so on) is a pass; 3 to 7 is partial. | Factual text is cited more often in answers; a slogan carries no information worth quoting. |
| Content is current (dated within the last year) | 7 | A date is present on the page or in the sitemap (dateModified, datePublished, article:modified_time, lastmod). If the newest date is within the last year it is a pass; if older, partial. | AI answers tend to choose newer sources, and a peer reviewed study showed that rankers are sensitive to dates. |
| Author and expertise visible | 5 | Person markup or an author link (rel="author", an author class) is present. | It shows the person behind the content and is part of how trust is assessed. |
| /llms.txt published | 1 | /llms.txt returns 200, is not HTML and is longer than 50 characters. | It does no harm, but no effect on visibility has been shown: a study of 300,000 domains found no link with citation, and the major bots do not request the file. That is why it earns only 1 point. |
What do the score levels mean?
Scores fall into five levels. The level boundaries are the same in both versions; the descriptions reflect the version 2 weights.
| Score | Level | Assessment shown in the report |
|---|---|---|
| 85–100 | Cited | Almost everything agents need to read and quote your site is in place. |
| 70–84 | Readable | The foundation is solid; close the few remaining gaps and your content is ready to be quoted. |
| 50–69 | Visible | Your site can be read, but gaps in access or content make it hard to be quoted. |
| 30–49 | Emerging | Agents reach your site only in part; there are a few clear gaps in access or content. |
| 0–29 | Invisible | AI agents most likely cannot read your site, or your page does not appear in the index. |
A level is not a promise. A site in the “Cited” level has met the technical and content prerequisites; whether it is actually mentioned in answers depends on the strength of competitors, the type of question and the brand’s footprint on other sites. If one of the heavy checks in area A fails, the score drops quickly. That is a deliberate choice: however good its content, a site with a closed door cannot be cited.
Version history: from version 1 to version 2
Every report keeps the version it was scored with. Older reports are not rescored; they are shown with the weights of their own version. That way, when you compare a past score with today’s, you know what changed.
| Area | Version 1 | Version 2 |
|---|---|---|
| Access (in version 2: access and indexing) | 30 | 47 |
| Structured data | 30 | 12 |
| Page signals | 15 | 11 |
| Visibility signals (in version 2: citation-ready content) | 17 | 30 |
| Measurement | 8 | Not scored |
Version 1 was used until 26 October 2026. In September 2026 we removed two checks: whether bots are named in robots.txt (what matters is whether they are allowed) and page speed (weak evidence as a citation factor). Their points were redistributed within the same area.
Version 2 went live on 26 October 2026. On 2 November 2026 we reviewed the sources again and confirmed the weights without changes. Three findings drove the changes. First, measurements showed no direct effect of structured data on AI citation, so its weight dropped from 30 to 12. Second, Google’s only official requirement is to be indexed and eligible for snippets; the check that measures this entered the score with 12 points, and the access area rose to 47. Third, data shows that citations come from the top of the page and from current content; the opening paragraph check was added with 10 points, freshness rose from 4 to 7, and llms.txt fell from 4 to 1. The measurement checks left the score and moved to a separate box.
Limits: what this score does not tell you
To read this score honestly, there are a few limits you should know about.
- Home page centered: Most checks look at the home page. Your service pages may be better or worse than your home page.
- Rules approximate: The opening paragraph and fact checks scan text with rules. They may miss a definition sentence or count a slogan as a fact. We show what we found in the evidence column, so you can check it yourself.
- Testing from outside: Your firewall may stop our request while letting real bots through, or the reverse. In that case we record a partial result and say why.
- Evidence changes: Most studies of AI search come from vendor data and can go out of date within months. We review the weights as new evidence arrives and record every change in this section with a version number.
- A high score is not a guarantee: 100 points means the technical barriers to being cited have been removed. For your brand to appear in answers, you also need mentions on other sites, genuine expertise and content that beats your competitors.
Sources behind the scoring
All sources were accessed on 1 November 2026; studies reviewed in our internal research on 26 October 2026 were verified again.
- Google Search Central, AI features and your website (official).
- Google Search Central, Google-Extended and common crawlers (official).
- OpenAI, Overview of OpenAI crawlers (official).
- Perplexity, Perplexity crawlers (official).
- Search Engine Land, Microsoft Bing and Copilot use schema for their LLMs (March 2025).
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 (peer reviewed).
- C-SEO Bench, NeurIPS 2025 (peer reviewed).
- Fang et al., Recency bias in LLM rankers, SIGIR-AP 2025 (peer reviewed).
- Kevin Indig, ChatGPT citations content study, Search Engine Land, February 2026.
- Ahrefs, Do AI assistants prefer to cite fresh content? (July 2025).
- Ahrefs, Schema and AI citations (2026).
- SE Ranking, Does llms.txt affect AI visibility? (300,000 domains).
- Vercel, The rise of the AI crawler (December 2024).
- Seobaz, Türkiye AI Overview research 2026 (in Turkish).
- Chen et al., Earned media and brand bias in AI search (2025, preprint).
Questions about AI visibility and the analysis.
Will the analysis change anything on my site?
No. We only read public addresses: the home page, robots.txt, llms.txt, the sitemap and up to 4 sample pages we pick from the sitemap. We take a screenshot of the home page and measure its speed with Google PageSpeed. We do not fill in any form or log in.
Do you ask AI any questions?
For the score we do not ask ChatGPT, Gemini, Perplexity or Claude questions about you; we visit your site with the identity of their agents and measure what the agent would see. We use a language model (OpenAI) to summarize your page and write suggestions; the score comes from the checks, not from the model. If you choose when e-mailing yourself the report, we ask ChatGPT, with web search, 5 questions a customer of yours might ask, once each and without naming you; it is a single sample, not a ranking. The report also has ready-made questions that open in your own account.
How do I recognize the analysis in my server logs?
We open your home page first with the identity of ChatGPT’s search bot (OAI-SearchBot), then with Claude’s search bot (Claude-SearchBot). If the door is closed to bot identities, we also open the page once as an ordinary browser. We request robots.txt, llms.txt and the sitemap with the identity that could open the home page.
How is the score calculated?
Each check in the four areas has a weight; together they add up to 100. Close to half of the score goes to access and indexing, a third to content that answers up front, with facts and a recent date; structured data and llms.txt get a small share because no direct effect on AI citations has been shown. Measurement setup (GA4, pixel) is not scored. A check that is fully met gets its whole weight, a partly met check gets half. We set these criteria ourselves and show them next to every result in the report; it is not a score published by AI companies.
Which pages do you look at?
The score of the 22 checks comes from the home page and the root files (robots.txt, llms.txt, sitemap); that is where an agent first gets to know your site. We also open up to 4 sample pages from page types in your sitemap, such as service, product or blog pages, and show them separately in the report; they do not count toward the score. Reviewing all your pages one by one and comparing you with competitors is the subject of the deep analysis.
Can I check again after making fixes?
Yes. The same report opens for the same site for twelve hours; once your fixes are live, the “I fixed it, measure again” button at the top of the report measures again without waiting. The new report is compared with the previous one: it shows which checks were fixed and which are still open. The same site can be measured again once every 10 minutes.
Are GEO analysis and AI visibility analysis the same thing?
They are two layers with the same goal. An AI visibility analysis measures whether your site is technically ready for AI agents. A GEO analysis measures, side by side with your competitors, whether ChatGPT, Gemini, Perplexity and Claude mention and cite you for the questions your audience asks.
Why doesn't ChatGPT recommend my brand?
The most common reasons: the OAI-SearchBot search bot is blocked in robots.txt or by the security layer, the service description depends on JavaScript and cannot be read, pages do not answer the question clearly, or other sources describe the brand rarely and inconsistently. A model cannot know what was published after its training ended; for current questions it searches the web and uses the pages it can reach.
What does it take to appear in Google AI Overviews?
According to Google there is no extra requirement or special optimization: the page has to be indexed and eligible to be shown in Search with a snippet. AI Overviews are built from these pages; being visible takes a solid SEO foundation and content that answers the question clearly.
Is an llms.txt file required?
No. llms.txt is not an official standard; it is an independent proposal published at llmstxt.org. Google states that Search does not use the file. It is cheap to set up, but it does not replace robots.txt rules or good content.
What happens if I block AI bots?
It depends on which bot you block. Blocking search bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot) can remove you from those answer engines' search results. Blocking training bots (GPTBot, ClaudeBot, the Google-Extended control) asks that your content not be used for model training, and it does not affect Google Search.
How do I see visits that come from AI?
In Google Analytics 4, referrals from chatgpt.com, gemini.google.com, perplexity.ai, copilot.microsoft.com and claude.ai can be tracked separately. Copilot and Bing citations appear in the AI Performance report in Bing Webmaster Tools. Traffic from Google's AI Overviews is part of overall Web traffic in Search Console and is not shown separately.
The panel our team watches, open for your brand too.
After the analysis: GEO and SEO consulting.
Related services
- GEO ConsultingContent, entity and measurement work to be mentioned and cited in AI answers.
- AI Access ControlSetting the door for AI agents with robots.txt, the security layer and llms.txt.
- Google AI Overview ConsultingThe search groundwork for visibility in Google AI Overviews and AI Mode.
- SEO ConsultingThe crawling, indexing and content base that AI visibility rests on.
- ChatGPT Ads ConsultingAdvertising in ChatGPT: account, targeting and conversion tracking.
From the blog
- What is GEO?What Generative Engine Optimization is and how it differs from SEO.
- What is Google AI Mode?How AI Mode differs from AI Overviews, and what it means for your site.
- What is robots.txt?The robots.txt file and how to allow AI bots.
- What is an LLM?How large language models work and what that means for brands.
The official sources behind this guide.
17 sources, 7 of them from Google.
| Source | Publisher | What it supports in the guide |
|---|---|---|
| 1AI features and your websitedevelopers.google.com/search/docs/appearance/ai-features | Google Search Central | No extra requirements; indexing and snippet eligibility; query fan-out; Search Console reporting; preview controls. |
| 2Optimizing your website for generative AI features on Google Searchdevelopers.google.com/search/docs/fundamentals/ai-optimization-guide | Google Search Central | That optimizing for generative AI is SEO; notes on llms.txt, chunking and artificial mentions. |
| 3List of Google's common crawlers (Google-Extended)developers.google.com/crawling/docs/crawlers-fetchers/google-common-crawlers | Google for Developers | That Google-Extended governs use in Gemini training and grounding and does not affect Search. |
| 4General structured data guidelinesdevelopers.google.com/search/docs/appearance/structured-data/sd-policies | Google Search Central | The rule that structured data must represent the content visible on the page. |
| 5About knowledge panelssupport.google.com/knowledgepanel/answer/9787176 | Google Knowledge Panel Help | Where Knowledge Graph facts come from; claiming the panel and suggesting changes. |
| 6Overview of OpenAI Crawlersdevelopers.openai.com/api/docs/bots | OpenAI | The roles of OAI-SearchBot, GPTBot and ChatGPT-User; robots.txt changes taking effect in about 24 hours. |
| 7Does Anthropic crawl data from the web, and how can site owners block the crawler?support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler | Anthropic (Claude Help Center) | The roles of ClaudeBot, Claude-SearchBot and Claude-User, and how blocking affects visibility. |
| 8Perplexity Crawlersdocs.perplexity.ai/docs/resources/perplexity-crawlers | Perplexity | That PerplexityBot serves search and is not used for training; that Perplexity-User generally does not follow robots.txt. |
| 9Bing Webmaster Guidelineswww.bing.com/webmasters/help/webmaster-guidelines-30fba23a | Microsoft Bing | Giving key information at the top of the page; URLs focused on a single topic. |
| 10Introducing AI Performance in Bing Webmaster Tools Public Previewblogs.bing.com/webmaster/2026/2/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview/ | Bing Webmaster Blog (Microsoft) | Measures of citations in Copilot and Bing AI summaries (February 10, 2026). |
| 11RFC 9309: Robots Exclusion Protocolwww.rfc-editor.org/rfc/rfc9309.html | IETF (RFC Editor) | The robots.txt standard; its rules are not a form of access authorization. |
| 12The /llms.txt filellmstxt.org/ | llmstxt.org (Jeremy Howard) | llms.txt defined as an independent proposal. |
| 13GEO: Generative Engine Optimizationarxiv.org/abs/2311.09735 | arXiv · KDD 2024 (Aggarwal et al.) | The finding that adding sources, quotations and statistics raised visibility by up to 40% on the test set. |
| 14Wikipedia: Conflict of interesten.wikipedia.org/wiki/Wikipedia:Conflict_of_interest | Wikipedia | The rule that organizations should not edit their own articles and that paid editing must be disclosed. |
| 15CCBotcommoncrawl.org/ccbot | Common Crawl | The crawler that builds the public web archive, and blocking it with robots.txt. |
| 16Web Vitalsweb.dev/articles/vitals | web.dev (Google) | The thresholds LCP 2.5 s, INP 200 ms and CLS 0.1. |
| 17Understand JavaScript SEO Basicsdevelopers.google.com/search/docs/crawling-indexing/javascript/javascript-seo-basics | Google Search Central | How Google crawls, renders and indexes JavaScript pages. |
No source matches this search.