AI visibility checker: A step-by-step method to measure your brand in AI search

---

title: "AI visibility checker: A step-by-step method to measure your brand in AI search"
description: "Learn how to use an AI visibility checker to test brand mentions, citations, crawler access, and competitive visibility across AI search engines."

AI visibility checker: A step-by-step method to measure your brand in AI search

If you cannot tell whether ChatGPT, Google AI Overviews, Claude, Perplexity, Microsoft Copilot, or Gemini can discover and describe your company, you need a repeatable test rather than a single score. An AI visibility checker should separate technical access, source retrieval, brand mentions, citations, and competitive presence.

This article is a method specification. It reports no TrustGrowth measurement, sample, or dated result. The 20-prompt worksheet later in the page is illustrative.

What follows is a directional method you can run with 10–20 prompts, repeat on a fixed schedule, and explain to your team. It does not produce an official Google or model-wide ranking. It shows how to record what happened for a defined prompt set, selected models, location, language, and test date.

Google AI Overviews are observed in Google Search result pages. ChatGPT, Claude, Perplexity, Microsoft Copilot, and Gemini are conversational systems. Record the observation protocol for each named engine instead of mixing those surfaces into one count.

What an AI visibility checker actually measures

AI visibility is the extent to which selected AI systems can discover, retrieve, represent, mention, and cite your brand for defined questions. A technical audit can show whether a page is reachable, but only an answer test can show whether a model actually includes your company.

Classic SEO visibility usually focuses on rankings, impressions, clicks, and indexed pages in Google Search. AI visibility adds answer inclusion, source retrieval, citation accuracy, and unprompted brand recall. These metrics can correlate, but they are not interchangeable. TrustGrowth's public write-up of why GEO belongs in a credibility score is the live source for that split.

Metric Question answered Evidence required Common misinterpretation Search visibility Does the site appear in Google Search? Google Search Console impressions, clicks, queries, and positions A high position guarantees AI citations Brand mention Does an AI answer name the company? Exact model response and test prompt A mention means the model endorses the company Citation Does the answer link to an owned or relevant source? Cited URL and claim-to-source comparison Any link is an accurate citation Retrieval Did the model use a relevant source? Source list, URL, or retrieval trace where available Retrieval guarantees recommendation Conversion Did visibility produce a business action? Analytics events, leads, trials, or revenue AI visibility alone caused the conversion

The first time you use TrustGrowth, treat its AI visibility diagnostics as one evidence source, not as a universal authority. A useful checker exposes its prompts, models, dates, sample size, and limitations.

Step 1: Define the visibility question before checking

Your result is only meaningful when you fix the product, category, audience, competitors, geography, language, models, and date range before testing. Changing “best analytics tool” to “best analytics tool for a technical founder” can change the retrieved sources and the answer.

Create two prompt groups. Branded prompts contain your company name; unbranded prompts describe the problem or category without naming your company. This branded versus unbranded split reveals whether the model recalls your company only after you provide its name.

Use at least 10–20 prompts for a directional sample. That sample is not a population-level measurement and cannot represent every user, wording, country, or model version.

Example prompts for an indie SaaS product include:

1. “What are the best [category] tools for technical founders?”
2. “How do I solve [problem] without hiring an agency?”
3. “Compare [brand] with [competitor].”
4. “Which [category] product shows evidence from Google Search Console?”
5. “What should a small product team check before publishing AI-search content?”

Record the exact wording, model, location, language, and date. ChatGPT, Perplexity, Google AI Overviews, Claude, Microsoft Copilot, and Gemini may produce different results from the same prompt, and the same model can change its answer on a later date.

Step 2: Check whether AI crawlers can access the site

A permission in robots.txt is not proof that an AI crawler fetched, indexed, retrieved, or cited your page. Start with representative pages: the homepage, one product page, one documentation page, one comparison page, and one author or company page.

Check these conditions:

• HTTP status is normally 200 for the intended page.
• Redirects resolve to the correct canonical URL.
• The page does not contain an unintended noindex directive.
• The XML sitemap lists the important URL.
• The server is available without authentication or intermittent failures.
• Key content renders in the delivered HTML, not only after blocked JavaScript execution.
• robots.txt permissions are reviewed for relevant crawlers.

Use Google’s robots.txt documentation, Google Search Console documentation, OpenAI’s crawler documentation, Anthropic’s web crawler documentation, and Perplexity’s crawler documentation as primary references.

Classify each page as reachable, blocked, or unknown. A reachable page passed your access checks. A blocked page failed because of permissions, status, rendering, authentication, or availability. Unknown means your test lacks evidence that the relevant crawler fetched or used it. This step-by-step method for checking AI crawler access explains why permission and observed access must remain separate.

Step 3: Run a repeatable AI answer and citation test

Run the same 10–20 prompts across your selected models and record the complete response, not only the first sentence. One ChatGPT response is an observation, not a stable ranking position.

What to record for every prompt

Field Example Prompt ID and intent U03, unbranded category comparison Model and date Perplexity, 2026-08-26 Brand mention Yes or no Citation Yes, with the exact URL Citation accuracy Supports the claim or does not support it Competitors Named companies and answer position Confidence and notes High, medium, or low; record variability

Use five visibility states: absent, mentioned without a link, linked as a citation, cited for a specific supported claim, or directly recommended. A brand mention is not equivalent to a citation.

Also record which source types appear repeatedly: official documentation, review sites, Reddit, G2, analyst pages, news coverage, or competitor pages. If competitors appear in 12 of 20 prompts and your brand appears in 3, compare the sources and claims behind those appearances instead of assuming that domain authority explains the difference.

Step 4: Calculate a transparent AI visibility snapshot

Calculate each dimension separately so your team can see what the measurement establishes and what it cannot establish. For a prompt set of 20, use these formulas:

• Mention rate = prompts with a brand mention ÷ total prompts.
• Citation rate = prompts citing a relevant owned page ÷ total prompts.
• Retrieval rate = prompts retrieving a relevant owned or third-party source ÷ total prompts.
• Competitive share of mentions = your brand mentions ÷ all tracked brand mentions.

Illustrative 20-prompt worksheet (not a measurement)

Metric Result What it establishes What it cannot establish Mention rate 6/20 = 30% The brand appeared in 6 tested answers Visibility for all prompts or users Citation rate 4/20 = 20% A relevant owned page was cited 4 times Citation quality outside the sample Retrieval rate 11/20 = 55% A relevant source appeared 11 times That the brand was recommended Competitive share 6/31 = 19% The brand supplied 6 of 31 tracked mentions Market share or future performance

This is an illustrative worksheet, not a TrustGrowth result or a claim about a real site. If you later publish a real run, publish the 20-prompt sample size, test date, models, locations, prompt set, and uncertainty with the percentage. Keep a sampled figure visually distinct from a census figure so a directional sample is not presented as a complete market census.

Do not report a bare “AI visibility score.” A transparent snapshot says what was tested and when. Re-run the identical test after a defined change to determine whether the result moved; improvements remain hypotheses until that later test shows movement.

Step 5: Diagnose why visibility is weak

Match the intervention to the failed dimension instead of publishing more content by default.

• Not reachable: fix crawlability, rendering, indexability, sitemap coverage, redirects, and server availability.
• Reachable but not retrieved: improve topic coverage, page structure, internal links, definitions, and authoritative supporting references.
• Retrieved but not cited: align claims with evidence, add unique data, clarify authorship, and link the source supporting each important claim.
• Cited inaccurately: make product, documentation, pricing, company, and author information consistent across the site and third-party profiles.
• Strong branded but weak unbranded visibility: publish problem- and category-oriented answers instead of more branded copy.
• Competitors dominate: identify the sources cited for competitors and fill missing evidence, third-party references, or answer coverage.

Record the failed test and the proposed fix. You do not know whether a content change caused improvement until the same prompts, models, and conditions are tested again.

Step 6: Turn checker results into a one-week action plan

Use this seven-day sequence to move from observation to a re-testable change:

1. Day 1: establish the 10–20 prompt baseline and capture exact outputs.
2. Day 2: verify crawler access, status codes, canonicals, noindex, rendering, and sitemap inclusion.
3. Days 3–4: repair the page or evidence gap you can re-test within the week.
4. Day 5: add concise answer sections, lists, comparison tables, definitions, and source links.
5. Day 6: validate structured data and authorship information. Schema markup does not guarantee a citation.
6. Day 7: rerun only the changed tests and record what remains unknown.

Prioritize each action by evidence strength, expected reach, effort, and ability to re-test. Use this checklist before publishing your result:

• Product, category, audience, competitors, geography, language, and date are fixed.
• Branded and unbranded prompts are separated.
• Each model and test date is recorded.
• Mention, retrieval, citation, and citation accuracy are separate fields.
• Competitor sources are captured.
• Reachable, blocked, and unknown pages are distinguished.
• Sample size and limitations are published.

Do not claim: a page is cited everywhere; an AI visibility score is universal; or robots.txt approval guarantees inclusion. None of those claims follows from a 10–20 prompt test or a permission check.

How TrustGrowth fits into the checking process

TrustGrowth connects to Google Search Console to verify owned-site search data and publishes GSC-verified site audits, scoring, and GEO diagnostics. Its GSC-verified site audit and scoring can validate site conditions and first-party search evidence available in the audit scope and date.

Public proof pages and the leaderboard let you inspect methodology, timestamps, and evidence instead of trusting an anonymous number. That boundary matters: a Google Search Console audit can verify your site and search data, but it cannot guarantee how a closed or changing AI model will answer every prompt.

Evaluate any checker by asking whether it reached your site, disclosed its prompt set, named its models, showed test dates, and exposed its limitations. TrustGrowth's live method for that check is how to tell whether an SEO tool measured your site or just failed to reach it.

AI visibility checker FAQ

What is an AI visibility checker?

An AI visibility checker is a method or tool that tests whether AI systems discover, retrieve, mention, represent, and cite a brand for selected prompts. Its quality depends on the prompt sample, models, test date, and evidence shown.

How do I check if my brand appears in ChatGPT or Google AI results?

Use a fixed set of branded and unbranded prompts, record exact outputs and citations from ChatGPT and Google AI features, and repeat the test over time. Distinguish a brand mention from a linked citation that accurately supports a specific claim.

Is AI visibility the same as SEO visibility?

AI visibility is not the same as SEO visibility. Classic SEO measures rankings, impressions, clicks, and related Google Search metrics, while AI visibility also measures retrieval, answer inclusion, mentions, and citations.

How accurate are free AI visibility checkers?

Free AI visibility checkers provide directional snapshots, but they may use undisclosed prompts, small samples, cached results, or one model. Choose a checker that discloses dates, models, prompts, sample sizes, and limitations.

Why does ChatGPT mention my competitor but not my company?

ChatGPT may mention a competitor because the competitor has stronger third-party evidence, clearer category relevance, better crawlability, or more frequently cited sources. Compare the exact sources and claims in the answers before changing your strategy.

Can robots.txt guarantee AI visibility?

No. robots.txt can permit or restrict access for some crawlers, but permission does not prove fetching, indexing, retrieval, mention, or citation.

Conclusion

This page specifies a checker method. It does not report a TrustGrowth measurement.

Fix the question, prompt set, models, and date first. Separate crawler access from retrieval, mention, citation, and citation accuracy. Treat Google AI Overviews as a search-page observation and conversational systems as chat observations. Publish the sample size and limits with any later real run. Re-test the same prompts after a change before claiming movement.

Method and limits

This article describes a repeatable directional method using a 10–20 prompt sample across selected AI systems, with each observation tied to a model and test date. The formulas are descriptive, not official Google, OpenAI, Anthropic, Perplexity, or Microsoft scores.

The method has two primary limits: a small prompt set cannot represent every query or user, and AI responses can change by model version, location, language, retrieval layer, and date. Technical access checks also cannot prove that a system fetched or cited a page. A later identical test is required to evaluate whether an intervention changed the observed result.

Start with 20 prompts, publish the worksheet, and make one repair you can test within 7 days. That gives you evidence you can inspect instead of a score you cannot explain.