Home/Insights/AI search guides
Insights / AI search visibility

How to check your AI visibility: a free checker and a manual method that holds up.

To check your AI visibility, ask Google AI Overviews, ChatGPT, Gemini and Perplexity the questions your buyers ask, and record for each answer whether your company was named, cited as a source or absent. Use a fixed list of questions, fresh sessions and a dated record, and the free checker on this page turns your marks into a score from 0 to 100. Below the checker, the guide covers how to read the result and how to turn the check into a proper audit and improvement plan.

01 / THE METHOD

How to check your AI visibility, step by step

A reliable AI visibility check takes eight buyer questions, four engines and a simple mark for every answer: named, cited only or absent. It costs nothing but time, and unlike most automated checkers you can see exactly what was asked and what came back. If you are checking because of headlines claiming search is finished, what marketers actually say about AI search in 2026 separates the fears from the evidence.

The manual AI visibility check
  1. 01 · QUESTIONSWrite up to eight questions in your buyers’ words, without your company name
  2. 02 · SESSIONSOpen fresh, signed-out or temporary sessions on each engine
  3. 03 · ASKAsk every question on every engine and save each answer
  4. 04 · MARKMark each answer: named, cited only or absent
  5. 05 · SCOREEnter the marks in the checker below for a 0–100 score
  6. 06 · REPEATRun the same questions again next month, on the same engines

Keep every raw answer with its date, so anyone can recount it.

Step 1: choose the questions

Use the questions a buyer asks before they know your name. Mix five types: category questions (“best … for …”), comparisons, requests for alternatives, problems and requirements such as a certification, a location or a budget. Leave your company name out; a question that names you tests whether the engine knows you, not whether it would recommend you. Our play on building a fixed question panel covers the wording in more detail.

Illustrative example · eight buyer questions

For a firm selling HR software to mid-sized UK manufacturers:

  1. Category: “What is the best HR software for a manufacturer with 300 staff?”
  2. Category: “Which HR systems handle shift patterns well?”
  3. Comparison: “All-in-one HR suite or separate payroll and HR tools: which is better for a factory?”
  4. Alternatives: “Alternatives to the HR system we outgrew, for a 300-person company.”
  5. Problem: “How do I stop holiday clashes on the shop floor?”
  6. Problem: “Our absence records are scattered across spreadsheets. What should we use?”
  7. Requirement: “HR software that is ISO 27001 certified and works with UK payroll.”
  8. Cost: “How much does HR software cost for 300 employees in the UK?”

Invented questions, written to show the mix. They are not real queries and name no company.

Step 2: set up the sessions

Answers depend on who is asking. Use a signed-out browser window or a temporary chat so your own history does not tilt the result, set the country your buyers are in, and note the date. Each engine works a little differently:

Running the check on each engine
EngineHow to run itWhat counts as citedWatch out for
Google AI OverviewsA normal Google search, signed out or in a private window, in the country your buyers are in.Your page appears among the links shown with the overview.Not every search shows an overview. If none appears, try AI Mode and note which one you used; Google says the two “may use different models and techniques”.
ChatGPTA new chat with web search on, or a temporary chat so your history does not shape the answer.Your page is among the sources or inline links.Without search, ChatGPT answers from its training data and shows no sources. Sites that block OAI-SearchBot are not shown in ChatGPT search answers.
GeminiA new conversation, ideally signed out or in a separate profile.Your page is among the sources or links the answer shows.Some answers show no sources at all. Mark named or absent from the text alone, and note it.
PerplexityA new thread with default settings.Your page is one of the numbered citations.Perplexity searches the web in real time, so results can move quickly after you change a page.

Steps 3 and 4: ask, save and mark

Ask each question once per engine, then save the full answer: a screenshot or a copy of the text and the source list. Give every answer one mark. Named means the answer text includes your company or product name, whether or not it also links to you. Cited only means one of your pages is among the sources but the text does not name you. Absent means neither. If you also want to track competitors, mark their names in the same answers; a score only means something next to theirs.

Step 5: score it

Enter the marks in the checker below. It uses the same arithmetic as our guide to how an AI visibility score is calculated: named counts as one point, cited only as half a point and absent as nothing. Each engine is scored out of 100 and the headline is the average of the engines you ran. That article explains why the half point is there and how to change it.

Worked example (invented marks): eight questions on two engines. ChatGPT: 2 named, 2 cited only, 4 absent, so (2 + 2 × ½) ÷ 8 × 100 = 37.5. Perplexity: 4 named, 2 cited only, 2 absent, so (4 + 2 × ½) ÷ 8 × 100 = 62.5. The headline is the average of the two: (37.5 + 62.5) ÷ 2 = 50 out of 100, reported with the range “37.5 to 62.5 by engine”. Enter the same marks in the checker and it shows the same figures.

Step 6: repeat before you react

One run is a snapshot, not a trend. Research by SparkToro and Gumshoe.ai, reported by Search Engine Journal on 30 January 2026, ran 2,961 prompts and found that ChatGPT and Google’s AI returned the same list of brands “less than 1% of the time across repeated runs of the same prompt”. Run the same questions several times, or again next month, before you treat a change as real. The play on repeated runs suggests a practical minimum.

02 / THE CHECKER

Free AI visibility checker

Enter your questions and your marks; the score, the range between engines and the questions where you were absent everywhere update as you go. It works on a phone, so you can mark answers as you run them.

AI visibility checker. Type up to eight buyer questions, run each one on each engine, and mark what you saw. Leave an engine as “Not run” if you did not test it. Nothing you enter is sent anywhere; it stays in this page.

AI visibility score: not calculated yet

Mark at least one answer as named, cited only or absent.

      Copy the text result into a dated note or spreadsheet with your saved answers. The score becomes useful the second time you run it, on the same questions.

      03 / THE RESULT

      What your result means and what to check next

      The headline score tells you how often you appear; the engine scores and the absent questions tell you why. Look at the lowest engine and the questions where you were absent everywhere first.

      If the checker shows this, check this first
      What you seeFirst thing to check
      Low on Google AI OverviewsWhether the pages you expect are indexed and eligible for a snippet: Google says a page must be both to be a supporting link. Search Console now reports appearances in generative AI features.
      Low on ChatGPTWhether your robots.txt allows OAI-SearchBot. OpenAI says sites opted out of it “will not be shown in ChatGPT search answers”. Our AI crawler log-file check shows how to confirm the bots actually visit.
      Low on PerplexityWhether PerplexityBot is allowed; Perplexity says it is designed “to surface and link websites in search results on Perplexity”. Then compare the sources it cites with yours.
      Low on GeminiWhich sources it names instead, and whether your company facts match across your site and profiles.
      Absent everywhere on category questionsThe third-party lists and directories the answers draw on. See why brands are missing from AI shortlists.
      Cited but rarely namedWhether your cited pages say plainly what you sell and who it is for.
      Named, but described wronglyThe source the engine is repeating. See when AI describes your company wrongly.

      There is no universal “good” score: it depends on how hard the questions are. Compare yourself with last month and with two or three competitors on the same questions. The wider work that moves the number is covered in our AI search visibility services.

      04 / TOOLS

      Free AI visibility checkers vs checking by hand

      Free AI visibility checkers are quick, but most choose the questions for you and show a score without the answers behind it. A manual check is slower and fully transparent. Use a free tool for a first look and the manual method for anything you report or act on.

      Before you trust any checker’s number, free or paid, ask six questions:

      • Which questions? Generated from your website, or your buyers’ real questions? Can you see and edit them?
      • Which engines? All four your buyers use, or one?
      • Interface or API? An answer from a developer API can differ from what a buyer sees in the app.
      • How many runs? One run per question is a snapshot, as Step 6 explains.
      • Which country? UK and US answers can name different suppliers.
      • What is the formula? If the tool will not say how named, cited and absent are counted, you cannot compare its score with anything.

      Our comparison of AI visibility tools covers the paid platforms for when you need tracking at scale, and the guide to reporting AI visibility to a board covers what to put beside the score.

      05 / QUESTIONS

      Questions about checking AI visibility

      Is there a free AI visibility checker?

      Yes. The checker on this page is free and scores your own questions across Google AI Overviews, ChatGPT, Gemini and Perplexity with a published formula. Several AI visibility tools also offer free one-off checks; read what each one counts before you act on its score.

      How often should I check my AI visibility?

      Monthly is enough for most companies, on the same questions and engines each time. AI answers vary from one run to the next, so a single check is a snapshot and a change only means something when it holds across repeated runs.

      Why do I get a different answer each time I ask?

      AI engines generate each answer afresh and can draw on different sources each time. Research reported in January 2026 found ChatGPT and Google’s AI returned the same brand list less than 1% of the time across repeated runs. How often you appear across many runs is the steadier measure.

      Does Google Search Console show AI Overviews?

      Yes. Google includes traffic from AI Overviews and AI Mode in the Performance report under the Web search type, and in June 2026 it introduced generative AI performance reports showing how often your URLs appeared in its generative AI features.

      Can I check a competitor’s AI visibility?

      Yes. Run the same questions and mark the competitor’s name in the same answers. Comparing your score with two or three competitors on the same panel is more useful than your score on its own.

      Sources for this method: Google Search Central, AI features and your website; Google Search Central blog, 3 June 2026; OpenAI, crawlers; Perplexity, crawlers; Perplexity Help Center, how Perplexity works; Search Engine Journal, 30 January 2026. All accessed 13 November 2026.

      06 / FIELD NOTES

      Start with the buying decision you want to influence

      The quick check above tells you where you stand. The rest of this guide turns it into an audit that leadership can act on, and that starts with a narrower question than “how visible are we?”: visible to whom, during which decision, for which service and in which market? A specialist named in a relevant supplier comparison may have a more valuable opportunity than a brand mentioned in a general explanation, even if the audience is smaller.

      Begin with one service and one buying situation: the organisation, the person researching, the problem that triggers the search and the constraints on the purchase. A chief operating officer considering workflow automation asks different questions from a marketing analyst learning a new term.

      The audit produces four outputs: a defined set of buyer questions, a dated record of observed answers, a prioritised content and technical backlog, and a measurement plan for business outcomes. It does not claim to measure the entire AI market or predict future answers, and saying so makes it easier for management to see what the evidence supports.

      07 / FIELD NOTES

      Define visibility before you create a score

      The checker’s three marks are deliberately simple. For an audit, keep four observations apart: a mention (the answer names the business), a citation (it links to your source), a recommendation (it presents you as suitable for the stated need) and a referral (someone follows a link to your site). A citation to a technical article may support authority without recommending you as a supplier.

      Record the context too: was the business described accurately, positively or with a limitation you should address? And keep branded and non-branded questions in separate views. Asking an assistant to explain your company tests whether it understands a business it has been told about; asking for help choosing a partner tests whether buyers would discover you. Better branded descriptions are not proof of discovery.

      Whatever you report, state the numerator, denominator, engines, period and question set: “named in 12 of 60 recorded answers” can be checked; “our AI visibility is 20%” cannot. Our guide to how AI visibility scores are calculated covers the arithmetic and how vendors’ formulas differ.

      08 / FIELD NOTES

      Build a question set from real commercial situations

      The eight questions in Step 1 are enough for a first look. For an audit, draw the wording from sales conversations, proposal objections, customer interviews and the questions that come up before a project is approved. These give you the buyer’s language rather than the service catalogue’s: a buyer may ask why paid leads do not become opportunities without knowing that the answer involves conversion tracking, landing pages and CRM feedback.

      Cover several stages: recognising the problem, comparing approaches, choosing a supplier, planning implementation and justifying the investment. Add realistic constraints such as industry, size, existing systems and location, but not so many that the question effectively names you. For each question, record the buyer role, the stage, the reason it belongs and the page you would want a serious buyer to find.

      Keep the core panel manageable. Ten to fifteen questions per service, the size our score guide uses, is a practical start; it is an operating choice, not a representative sample of demand. Have sales review the wording without seeing the expected result, and test emerging questions in a separate exploratory group until there is a reason to promote them.

      The operating sequence
      1. 01Define the decision
      2. 02Establish the baseline
      3. 03Implement the scope
      4. 04Review the outcome

      A working sequence, not a forecast of results.

      09 / FIELD NOTES

      Design a repeatable observation protocol

      Steps 2 to 6 above are the protocol in short form. Two additions make it hold up in an audit. First, if you test a multi-step buyer conversation, label it as a sequence and keep it apart from independent one-question runs: a follow-up carries context from the previous answer, so it does not show whether the engine would discover you unprompted.

      Second, keep the protocol stable between review periods. If you change the questions, the engines or the scoring rules, annotate the break and keep the old series; otherwise a rising score may reflect easier questions rather than better visibility. Record every repeated outcome, not only the best screenshot.

      10 / FIELD NOTES

      Read the cited sources as a competitive map

      When competitors appear, inspect the sources supporting the answer. The cited page may be the competitor’s own site, a directory, an industry publication, a partner page or a review platform. The practical question is what credible information the engine found there that it could not easily find about you.

      Group the cited sources by function: some explain the category, some compare providers, some demonstrate experience and some establish identity. A service page that says little beyond “we deliver results” is weaker evidence than a competitor page that explains scope, requirements, examples and measurement. Improve your evidence rather than imitating their phrasing.

      Check accuracy as well as presence. If you are linked to an outdated service or location, find where that information is still published and correct the authoritative source first; the audit should separate market positioning from incorrect or stale information. Then keep a source opportunity register: for each gap, the buyer question, the observed source, the missing information, an owner and an action.

      11 / FIELD NOTES

      Check whether your strongest pages can be found and understood

      Confirm that the pages that matter commercially load, can be crawled and are eligible for indexing; check canonicals, redirects and accidental noindex settings, and make sure the important explanation is in page text rather than only in an image. Google states that established SEO practices remain relevant to AI Overviews and AI Mode, with no special AI markup required, and that supporting pages must be indexed and eligible for snippets (Google Search Central: AI features and your website).

      Give important services a clear place in the site structure, linked from the services hub, from relevant guides and to their evidence, with descriptive link text. Treat structured data as a description of visible facts, not a substitute for a useful answer. Crawler access for ChatGPT and Perplexity is covered in the results table above.

      12 / FIELD NOTES

      Improve the pages that support a buying decision

      Begin each service page with the business problem and the outcome the work supports, then explain how the engagement runs, what it includes and what the client receives. A buyer should be able to tell whether the service fits without booking a call to decode it.

      Answer the questions an internal sponsor will face: what is examined, what data is needed, who implements, how progress is reviewed and what the work does not promise. Place evidence beside the claim it supports, with the measure and period labelled. Then give the reader a route to act: a focused enquiry or a relevant guide, with the service context kept attached to the request.

      13 / FIELD NOTES

      Separate observed referrals from the wider influence of AI

      In GA4, visits from ChatGPT, Gemini, Copilot and similar tools fall into the AI Assistant channel, while Google’s AI Overviews and AI Mode are counted in Organic Search, so check how your own set-up classifies them. Many AI-influenced journeys will still arrive as brand searches, copied URLs or later direct visits. Do not turn missing attribution into a precise estimate of invisible AI influence.

      Keep three views: observed answer visibility, observed website behaviour and CRM outcomes. They inform one another without proving one cause. An optional “What prompted you to contact us?” question in qualification adds self-reported context; keep it alongside the measured source rather than overwriting it, and report “source unavailable” openly.

      14 / FIELD NOTES

      Prioritise the work by commercial relevance and evidence

      Rate each backlog item on four things: the importance of the buyer question, the strength of the observed gap, the feasibility of the fix and its usefulness to the sales journey. Use simple ratings with written reasoning. A broken service page with active demand normally comes before an experimental format with no clear buyer use.

      Separate foundational repairs from experiments, because they are reviewed differently: a technical fault can be verified directly, while a change in how engines describe you may take longer and stay variable. Give every content item a business owner who confirms scope and accuracy, so the programme stays tied to what the company actually delivers.

      15 / FIELD NOTES

      Use a six-week audit and implementation sprint

      One way to turn the check into an audit with changes behind it:

      A six-week audit and implementation sprint
      WeekWhat happens
      1Agree the service, buyer situations and question panel; collect current pages and evidence; record the baseline before anything changes.
      2Run the protocol; review cited sources; note recurring gaps and wrong descriptions.
      3Inspect the site foundations; choose a small, ordered set of changes with owners.
      4–5Implement and verify: mobile, internal links, forms that keep the service context, accurate evidence captions, the enquiry path tested end to end.
      6Repeat the stable panel; compare with website and CRM evidence; agree the next cycle.

      Some changes will not yet show in AI answers by week six; report that honestly. The timetable is an operating structure, not a promise about when an engine will cite a page.

      16 / FIELD NOTES

      Create a board-ready visibility brief

      Keep the leadership brief short and link it to the evidence: the buying decisions you are trying to influence, the audit scope, the most consequential finding, the actions taken and the next decision. Use a small set of indicators with explicit definitions, each with its sample and period, and explain uncertainty in terms of which conclusions it does and does not allow. End with a funded next scope and a review point. A simple record for each decision:

      A working decision record
      RecordWhat to write
      ObservationWhat happened, in which period and sample?
      EvidenceWhich source supports the observation?
      AlternativeWhat else could explain the result?
      DecisionWhat will change, and why?
      OwnerWho implements and verifies it?
      ReviewWhen will the next decision be made?
      17 / FIELD NOTES

      A working example for an implementation partner

      Consider a hypothetical company that designs operational software for multi-location businesses and wants more qualified implementation enquiries. Its panel covers problem diagnosis, build-versus-buy comparisons, integration planning and supplier evaluation. A branded question produces a reasonable description of the company, but non-branded questions mostly cite generic software directories. This is an illustrative scenario, not a client result.

      The source review finds that the homepage talks about innovation, but the service pages do not explain delivery scope, supported integrations or the evidence behind the company’s experience. The first priority is one service page and one implementation guide: a clear account of discovery, integration responsibilities, testing and handover, a properly scoped case study and a service-specific enquiry form.

      A later run shows several new citations, which the team does not report as revenue. Sales finds two relevant conversations that mentioned the guide, one from a measurable referral and one from a colleague’s recommendation, and the report keeps both sources rather than forcing them into one AI category. The next decision is to strengthen the integration evidence and keep the same panel for another cycle.

      18 / FIELD NOTES

      Score an answer without overstating precision

      The checker’s three marks are quick to apply. For an audit, add an accuracy check on how the business is described and a relevance check on the cited page, and have two people mark a subset of answers independently. If they disagree on what counts as named, the definitions need tightening before the score goes to management. When you summarise, show a few representative answers beside the numbers: a strong one, an absence on an important question and any inaccurate description.

      19 / FIELD NOTES

      Turn one finding into an implementation brief

      Suppose a priority service is rarely cited because its page does not explain implementation responsibilities. Write the brief around that missing information: the buyer’s question, what the page needs to say, the subject-matter owner, the evidence available, the sections to change and the guide or case study to link.

      Add an acceptance check. The content owner confirms accuracy and scope, the website owner verifies the page, links and form, the measurement owner records the release date, and the audit owner schedules a comparable run. Then define the commercial question for the next review, such as whether the page attracts more relevant enquiries. That is how AI visibility work earns a durable place in a commercial strategy.

      20 / FIELD NOTES

      References and audit boundaries

      The question-panel method, sprint structure and leadership brief are proposed working tools. They do not represent a validated market-wide visibility index or a guarantee of inclusion. For platform foundations, see Google’s AI search guidance and helpful content guidance. For website measurement, see Google Analytics channel definitions.

      Keep a copy of the source documentation used during implementation and review material platform changes before the next audit. Separate current platform guidance from the organisation’s own operating choices.

      Read next.

      Keep the whole journey connected.

      Your next step

      Let’s make the next step a useful one.

      Tell us what is happening today and what you want to change. We’ll help you identify the right starting point.

      More than a decade of experience, across businesses and markets.

      Explore our work first