What does an AI brand visibility checker actually check?

Most checkers follow the same loop, whatever they call it.

1. Pick buyer questions. These are prompts a customer might ask, such as "best payroll software for a 20-person agency". Some tools let you write them. Others draft them from your homepage.
2. Ask each engine. The checker sends every prompt to each AI platform it covers, such as ChatGPT, Gemini, Perplexity or Claude.
3. Record mentions. For each answer it notes whether your brand is named, whether competitors are named, and which URLs are cited.
4. Check facts. If the answer states your price, founding year or product names, the checker compares those claims with facts you supplied. A wrong price is a finding. A vague but true line is not.
5. Score it. The results roll up into a number, per platform or blended across platforms.

That is the whole job. A checker observes what engines say. It does not change what they say. The accuracy part matters more than it looks: AI platforms get specific facts wrong about specific brands, and a check that only counts mentions will miss a wrong price entirely. A pricing accuracy check shows that failure on its own.

A visibility score summarises a small, specific sample. Read it with three limits in mind.

It is sampled

The score covers only the prompts that were asked. Three prompts and fifty prompts describe different things. Change the prompt list and the score can move with nothing else changing.

It mixes retrieval states

Engines decide per question whether to search the web. When they don't, the answer comes from training data frozen at a cutoff date, and nothing you publish this quarter reaches it. Any single score averages answers given with and without live retrieval. That includes ours.

It is not a rank

Google's Search Console reports AI Overview and AI Mode impressions only, with no position. No engine publishes position data, so any "AI rank" has nothing to be validated against.

Per-platform scores hide less than one blended number. Google states that AI Overviews and AI Mode "may use different models and techniques", so averaging them assumes a shared scale that doesn't exist. The trade-offs are laid out in one score vs per-platform scores, and what a visibility score means walks through reading one.

Run a check twice and the answers can differ. That is how the engines work, not a bug in the checker.

• Queries get rewritten. As of October 2026, OpenAI's help centre says ChatGPT search rewrites your question into one or more targeted queries before searching. The checker sees your prompt, not those queries.
• Retrieval is optional. ChatGPT searches at the model's discretion. Claude.ai search is a user toggle. Google AI Overviews often don't trigger at all.
• Context changes answers. OpenAI says saved memories can shape how ChatGPT rewrites a search. A reproducible check runs logged out with memory off.
• Models change. Vendors swap default models without notice, so a result from last quarter may describe a different system.

What it means: treat one run as a reading, not a verdict. Compare results only when the prompts, engines and settings match.

Be suspicious of any tool that claims to see these:

• The questions real buyers ask. No consumer AI app exposes user prompts or the queries it ran. Bing Webmaster Tools shows grounding queries for Copilot, in its interface only.
• Full citation volume. In Otterly's 2026 experiment, Microsoft's own AI Performance report credited one test site with 42,817 citations, while Otterly's prompt-based tracker logged 3,934. The time windows may not match, but it shows that a prompt sample sees a fraction.
• Influence. Being cited is not the same as being used. Research on how models cite (arXiv 2606.28358) repaired over 90% of missed citations by adjusting internal model components without changing the sources.
• Authority scores. Engines don't compute E-E-A-T scores or domain authority, so a checker reporting one invented it.

Coverage gaps also exist across platforms. Few tools cover every surface, and it's worth asking whether one tracks Meta AI before you compare scores.

A result tells you where to look. Here's the order that follows the evidence.

If you are not named

Check access first, because those levers are documented by the engines themselves.

• Confirm robots.txt allows each retrieval bot: OAI-SearchBot for ChatGPT search, PerplexityBot, Claude-SearchBot. Blocking GPTBot only stops training use. OpenAI says each setting is independent.
• Check that your firewall or CDN doesn't return 403 to crawlers that robots.txt allows.
• Make sure key content is in the HTML before JavaScript runs. Most AI crawlers don't execute it.
• Confirm your pages are indexed in Google and Bing, and that noindex or nosnippet aren't applied by accident.

When engines cite other sites instead of yours, being named on those sites is associated with appearing in answers.

If the facts are wrong

Find contradictions on your own site first: the price on the pricing page, the footer and the about page should match. State specific facts plainly on fetchable pages.

If you want a trend

A one-off check is a snapshot. Repeated runs on the same prompts show change over time, which is the core of the checker vs tracker distinction. The free checker on this site runs up to three buyer questions and scores them as a starting snapshot.

Is an AEO visibility checker different from an AI brand checker? Usually not. AEO (answer engine optimization), GEO and AI visibility checkers all ask AI engines questions and record whether a brand appears. Differences are in the details: which engines are covered, how prompts are chosen, whether facts are checked for accuracy, and whether scores are reported per platform or blended. Compare those details rather than the label on the tool.

Can a checker guarantee my brand will appear in ChatGPT? No. A checker observes answers. It does not control them, and no one outside the AI companies decides which sources an engine cites. What you control is whether crawlers can reach your pages, whether content renders without JavaScript, and whether your facts are consistent. Fixing those removes blocks. It does not promise a mention or a score.

Why does my score change when I haven't changed anything? Engines decide per question whether to search the web, rewrite prompts into their own queries, and update models without notice. Your score can also change if the checker's prompt list or engine coverage changes. Compare runs only when prompts, engines and settings match, and treat small swings between single runs as noise rather than a trend.

Does adding schema markup improve my checker result? The evidence doesn't support that. Google states structured data isn't required for generative AI search. Ahrefs' matched-control test of 1,885 pages found AI Overview citations fell 4.6% after schema was added. Existing markup does no harm, but it isn't a lever for AI answers. Crawler access and server-side rendering have far stronger evidence.

How many prompts does a useful check need? There's no published threshold. More prompts reduce the effect of any single odd answer, but they still sample questions you chose, not the queries engines actually run. Start with the questions that matter most to buyers, such as category comparisons and pricing, and keep the same set when you re-run so results stay comparable.