You want to run an AI visibility check without buying another dashboard first.
Fair enough. You can learn a surprising amount with a spreadsheet, a browser and five focused tests.
This guide shows you what to ask, what to record and how to turn scattered AI answers into a useful baseline.
A useful AI visibility check measures more than your brand name
AI visibility is not one ranking. ChatGPT may mention you while Gemini does not. Perplexity may cite your website for an informational question but recommend three competitors when the prompt becomes commercial.
That is why a one-off question such as “Do you know Company X?” tells you very little. It tests whether an assistant can retrieve or recall your name, not whether potential customers will encounter you during research.
For a practical baseline, choose 10 prompts that represent real stages of a buying journey:
- Three problem prompts: “How can a mid-sized manufacturer reduce energy costs?”
- Three category prompts: “Which energy consultants work with manufacturers in Germany?”
- Two comparison prompts: “Company A vs Company B for industrial energy consulting”
- Two trust prompts: “Is Company A a credible energy consultancy?”
Run the same prompts in the AI products your customers are likely to use. Record the date, answer, brand position, cited sources and competitors mentioned. Do not quietly rewrite a prompt until you appear.
The goal is not to produce a flattering screenshot, but to measure whether your brand enters realistic customer conversations.
This is the practical side of GEO vs. SEO: SEO measures visibility in search results, while GEO (Generative Engine Optimization) also examines how brands are represented inside generated answers.
Test 1 reveals whether AI systems understand your brand at all
Start with direct questions about your company. Use your exact brand name, common spelling variations and, if necessary, the combination of brand plus location or product category.
Ask questions such as:
- “What does Company X do?”
- “Who is Company X for?”
- “What products or services does Company X offer?”
- “Where does Company X operate?”
Score each answer from 0 to 2. Give 0 points if the system does not know you or confuses you with another entity, 1 point for a broadly correct but thin answer, and 2 points for an accurate description of your offer and audience.
If an assistant cannot connect your name to the right entity, category and market, broader recommendation visibility is unlikely.
Watch for confident errors. A detailed answer with the wrong services is worse than “I don’t have enough information,” because it can mislead a prospective customer. Save the wording rather than recording only a pass or fail.
Test 2 shows whether you appear before someone knows your name
Direct recognition is the easy test. Commercial discovery matters more.
Now remove your brand from the prompt and ask the assistant to recommend providers, products or approaches in your category. Include genuine constraints such as company size, use case, industry and geography.
For example: “Which B2B SEO consultancies support mid-sized companies in the DACH region?” is more useful than “What are the best SEO agencies?” The narrower version is closer to an actual brief and makes the comparison set meaningful.
Run each prompt in a fresh conversation. Previous messages can steer later answers and contaminate the test. If the interface offers web search, record whether it was enabled rather than mixing searched and non-searched responses.
Your real discovery score is the share of relevant, unbranded prompts in which your company appears.
If you appear in 3 of 10 unbranded prompts, your simple mention rate is 30%. The number is illustrative, but the logic holds. Repeat the same fixed set later and you have a baseline against which change can be measured.
Test 3 separates a passing mention from a supported recommendation
A brand mention can be little more than a name in a long list. A cited, accurately explained recommendation is stronger.
For every answer, capture four things: whether you were named, how prominently you appeared, what claim was made and which source supported it. Then open the cited pages. A citation beside an answer does not necessarily support every sentence in that answer.
You may discover that AI systems cite your homepage for your own services but rely on directories, editorial comparisons or specialist publications when recommending vendors. That source gap tells you where your website alone is not enough.
A useful citation is one that supports the right claim about your brand in the context of the customer’s question.
This is also why earning contextual brand mentions for GEO and SEO matters. You are not chasing mentions as decoration. You are building independent evidence that connects your company with a topic, use case and market.
Test 4 uses competitors to expose the evidence you are missing
Repeat your category and comparison prompts, then list every competitor that appears. Count mentions and citations instead of relying on which names feel most prominent.
Suppose 10 prompts produce 30 total company mentions. Competitor A receives 8, Competitor B receives 6 and you receive 2. Your share of mentions is 2 divided by 30, or 6.7%. This is not a universal industry metric. It is a controlled comparison within your own prompt set.
Next, examine why the leaders may be easier to recommend. Do they have clearer product pages? More third-party reviews? Better comparison content? Original research? Consistent descriptions across trusted sources?
Competitor visibility becomes actionable only when you connect each mention to the evidence that may have earned it.
Do not copy a competitor’s content plan blindly. The point is to identify patterns in the sources AI systems repeatedly use, then decide which gaps are commercially relevant to you.
Test 5 checks whether your visibility survives normal variation
Generated answers vary. Wording, citations and recommendations can change between platforms, sessions and dates.
Take your five most important prompts and run them across at least three relevant AI experiences. Then repeat the test on another day using the exact same wording. That gives you 30 observations: 5 prompts multiplied by 3 platforms and 2 dates.
If your brand appears once, you have an anecdote. If it appears in 20 of 30 observations with accurate positioning, you have a more credible signal. Record incorrect appearances separately so that visibility does not hide a quality problem.
Stable AI visibility means being mentioned accurately across repeated, commercially relevant prompts, not winning one carefully selected query.
Location, language, account settings and web access can affect results. Document what you can, and resist false precision. A manual test is a sample, not a census of everything every potential customer sees.
A simple scorecard turns the five tests into a baseline
Use one spreadsheet row per prompt and platform. Add columns for prompt type, brand mentioned, position, description accuracy, citation present, citation relevance, competitors and test date.
A compact 10-point score for each observation can work like this:
- Brand mention: 0 or 2 points
- Top-three placement in a list: 0 or 2 points
- Accurate description: 0 to 2 points
- Supporting citation: 0 or 2 points
- Citation actually supports the claim: 0 or 2 points
If 30 observations produce 174 points out of a possible 300, your baseline is 58%. Keep the raw fields, though. A single score can conceal the difference between poor recognition, weak citations and inaccurate answers.
The score is useful because it makes repeat tests comparable, not because 58% has a universal meaning.
Run the same set monthly or quarterly, depending on how actively you publish, improve pages and earn external coverage. Changing half the prompts each time destroys the comparison.
Your free check should end with a short action list
Look for the weakest stage rather than trying to “optimize for AI” in the abstract.
If direct recognition fails, clarify who you are, what you offer and where you operate on your own site and relevant profiles. If unbranded discovery fails, build useful category and use-case content. If citations are weak, improve source-worthy pages and pursue credible third-party coverage. If answers are inaccurate, find the source of the wrong claim and correct what you control.
Strong technical and content foundations still matter. The choice is not AI visibility OR SEO, and the work does not automatically require an endless contract. The distinction between a defined diagnostic and ongoing improvement is similar to one-time SEO versus a monthly retainer.
A free manual check tells you where the problem sits; a proper GEO audit is useful when you need broader prompt coverage, source analysis and a prioritised plan.
Start small: 10 prompts, three platforms, one documented baseline. You will know far more than you do after typing your brand into one chatbot and hoping for a pleasing answer.
Common questions about AI visibility checks
What is an AI visibility check?
An AI visibility check measures whether systems such as ChatGPT, Gemini, Perplexity or Google’s AI experiences mention and accurately describe your brand for relevant prompts. A useful check covers direct brand questions, unbranded category searches, recommendations, competitors and citations. It should use a fixed prompt set so that you can compare results over time.
Can I run an AI visibility check for free?
Yes. You can create 10 realistic customer prompts, run them manually across the AI tools available to you and record mentions, accuracy, citations and competitors in a spreadsheet. Free checks provide a useful baseline, but they remain a sample. They do not capture every user context, location, model version or answer variation.
What is an AI visibility score?
An AI visibility score is a summary of performance across selected prompts and platforms. It may combine mention rate, answer prominence, accuracy and citation quality. There is no single universal score, so the methodology matters more than the headline number. Keep the underlying observations and use the same method whenever you repeat the check.
How often should I check my AI visibility?
Monthly testing makes sense when you are actively publishing content, improving key pages or earning third-party mentions. Quarterly checks may be enough for a slower programme. Use the same core prompts, platforms and scoring rules each time. Otherwise, a higher score may reflect a changed test rather than genuinely improved visibility.
What is the difference between an AI visibility checker and an AI visibility audit?
A checker usually gives a quick snapshot of brand mentions for a limited set of prompts. An audit goes further by examining prompt coverage, competitors, answer accuracy, cited sources, content gaps and priorities for improvement. The checker tells you whether a problem may exist. The audit is intended to explain why it exists and what to do next.