AI visibility checker: how to check your AI visibility yourself
Most AI visibility checkers run one prompt once and show you a score. Here is the manual check that actually tells you something, in about an hour, with no tool.
The short answer
Run ten category questions across ChatGPT, Gemini, Claude, Perplexity and Copilot, three times each, recording whether you were named, whether you were linked, and which sources were cited. An hour, and better than most instant checkers.
Why instant checkers mislead
Most free AI visibility checkers run a small number of prompts once and return a score. Because model outputs are non-deterministic, that score would come out differently if the same tool ran again five minutes later, and nothing about your site would have changed.
The second problem is what they count. A tool reporting that you appear in 40% of answers is usually counting any mention of your brand name, linked or not. The stricter measure, whether a page on your domain was actually cited, is typically about half that. Neither number is wrong; presenting the first as citation rate is misleading.
The third problem is that a score with no source data attached tells you nothing about what to do. The actionable finding is almost always which domains are being cited instead of you, and that is the field most checkers do not show. Our assistant citation benchmark publishes that question openly.
The manual check, in about an hour
- Write ten questions a buyer in your category would actually ask. No brand names in at least seven of them.
- Open ChatGPT, Gemini, Claude, Perplexity and Copilot. Use a logged-out or fresh session so personalisation does not flatter you.
- Ask each question on each platform, three times, starting a new conversation each time.
- For every answer, record three things: were you named, was your domain linked, and which sources were cited in order.
- Add one follow-up turn on at least three of them, narrowing the question the way a buyer would. Note whether you survive to the second answer.
- Total the columns. You now have a rough citation rate, a rough mention rate, and, most usefully, a ranked list of the domains that own citations in your category.
How to read what comes back
| What you see | What it means | Where the work is |
|---|---|---|
| Not named anywhere | You are not in the retrieved set, and possibly not a resolvable entity to the model | Crawler directives first, then off-site citation sources |
| Named, never linked | The model knows you but is sourcing the claim from someone else | The third-party pages describing you, and your own extractability |
| Named and linked on branded queries only | You are findable but not in the consideration set for category questions | Off-site presence on comparison and review sources |
| Named, and described wrongly | Usually one stale third-party listing propagating everywhere | Fix the listing. This is days of work, not quarters |
When a tool is worth buying
The manual check does not scale. Thirty prompts, five runs and five platforms is 750 queries a month, and doing that by hand is a job rather than a task. That is the point at which tooling earns its cost.
What you are buying is coverage and cadence, not judgement. Check what any tool counts before trusting its headline: whether citation rate includes unlinked mentions, whether platforms are averaged into a single score, and whether it will show you the underlying source data. Our measurement standard sets out the definitions to hold them to.
We use Profound, Ahrefs Brand Radar and WriteWorks, and we run them against our own published definitions rather than accepting theirs. WriteWorks is built by Surge45, which we say wherever we mention it.
Our position
Do the manual check first, even if you intend to buy a tool. An hour spent reading the actual answers about your category is worth more than a quarter of dashboards, and it makes you a far better judge of which tool is telling you the truth.
Frequently asked questions
Why does the same question give different answers?
Model outputs are non-deterministic by design. This is exactly why a single run, and therefore most instant checkers, cannot tell you much. Run each prompt several times and report the rate, as set out in how to monitor AI search visibility. OpenAI documents its own crawlers if you also want to check what is fetching your pages.
Should I use a logged-in account for the check?
No. Personalisation and memory will flatter you, because the model may have context about your interests. Use a fresh or logged-out session.
Is a free checker better than nothing?
Marginally, if you treat the output as a prompt to look closer rather than as a measurement. The risk is a team acting on a score that would have come out differently on a second run.
Sources
- Anthropic, does Anthropic crawl the web and how can site owners control it. Anthropic's own crawler documentation.
About Surge45 Team
AI Search & GEO Specialists
Surge45 is the digital discovery and growth strategic advisory for SaaS. We help software companies become the answer across Google, AI search and communities, then turn discovery into pipeline. We also build WriteWorks, our content engineering platform for AI search.
Get More SaaS Marketing Insights
Join 10,000+ SaaS marketing leaders who receive our weekly newsletter with actionable strategies.
Subscribe to Newsletter