How AI Visibility Checker Pro works: prompts, engines and scoring
One scan at a glance
- Engines
- 6: ChatGPT, Claude, Gemini, Perplexity, AI Overviews, AI Mode
- Buyer prompts
- 3, unbranded, scored
- Brand prompt
- 1, on ChatGPT and Perplexity, not scored
- Answers per scan
- 20
- Time
- about 1–2 minutes
- Output
- 0–100 score, per-engine results, full answers and sources
The short version
Step 1: You tell us who you are
Your website, brand name and what you sell. Optionally who you sell to, up to 3 competitors, and a country.Step 2: We write buyer prompts
3 unbranded buyer-intent prompts and 1 brand prompt, from fixed templates you can read below.Step 3: We ask 6 AI surfaces
20 live answers, collected in parallel in about 1–2 minutes.Step 4: We detect and score
Mentions, rank, citations and competitors, rolled into a 0–100 AI Visibility Score. The full answers unlock with your email.
1. Prompts
The exact prompts we send
- P1Buyer prompt · scored
What are the best {category} for {audience}? Give me a short ranked list with one line on why for each.
Example
“What are the best expense management software for seed-stage SaaS startups? Give me a short ranked list with one line on why for each.”
- P2Buyer prompt · scored
I'm looking for a {category}. Which companies or products would you recommend, and why?
“a” becomes “an” when your category starts with a vowel.
Example
“I'm looking for an expense management software. Which companies or products would you recommend, and why?”
- P3Buyer prompt · scored
If you list a competitorWhat are the best alternatives to {competitor} for {audience}?
If you don'tWhich {category} do experts and users trust most right now? List the top options.
Example
“What are the best alternatives to Fernpay for seed-stage SaaS startups?”
- P4Brand prompt · not scored
What is {brand} ({domain})? What do they do, and are they a good choice for {audience} looking for {category}?
Checks whether the engine knows you and how it describes you.
Example
“What is Ledgerly (ledgerly.example)? What do they do, and are they a good choice for seed-stage SaaS startups looking for expense management software?”
Before your inputs go into a template we strip control and invisible characters, collapse whitespace and cap lengths (brand 60 characters, category and audience 80, each competitor 60). No prompt can exceed 500 characters. If you leave the audience empty, we use “startups”.
20 answers per scan
Every buyer prompt runs on all 6 engines. The brand prompt runs on ChatGPT and Perplexity to keep each scan affordable. All 20 requests run in parallel, which is why a scan takes about 1–2 minutes rather than half an hour.
| Engine | P1Buyer | P2Buyer | P3Buyer | P4Brand |
|---|---|---|---|---|
| ChatGPT | Runs | Runs | Runs | Runs |
| Claude | Runs | Runs | Runs | Not run |
| Gemini | Runs | Runs | Runs | Not run |
| Perplexity | Runs | Runs | Runs | Runs |
| AI Overviews | Runs | Runs | Runs | Not run |
| AI Mode | Runs | Runs | Runs | Not run |
| Answers per scan | 20 | |||
2. Engines
The six engines, and how we query each
| Engine | How we query it | Prompts asked |
|---|---|---|
| ChatGPTOpenAI | How we query itA live answer captured from chatgpt.com with web search on (DataForSEO LLM Scraper). Falls back to the OpenAI API with web search if the capture fails.Live app capture, API fallback | Prompts asked3 buyer prompts + the brand prompt4 answers per scan |
| ClaudeAnthropic | How we query itAnthropic's Claude model queried through the API (DataForSEO LLM Responses).Model API | Prompts asked3 buyer prompts3 answers per scan |
| GeminiGoogle | How we query itA live answer captured from the Gemini app (DataForSEO LLM Scraper). Falls back to the Gemini API with Google Search grounding.Live app capture, API fallback | Prompts asked3 buyer prompts3 answers per scan |
| PerplexityPerplexity AI | How we query itPerplexity's Sonar model, which searches the web by default (DataForSEO LLM Responses).Model API (searches the web) | Prompts asked3 buyer prompts + the brand prompt4 answers per scan |
| Google AI OverviewsGoogle | How we query itA live Google search results page for the prompt, including the AI Overview block and its cited sources (DataForSEO SERP API). Many searches show no AI Overview; those are marked “no AI answer” and excluded from the score.Live Google results page | Prompts asked3 buyer prompts3 answers per scan |
| Google AI ModeGoogle | How we query itA live Google AI Mode answer and its sources (DataForSEO SERP API).Live Google AI Mode | Prompts asked3 buyer prompts3 answers per scan |
3. Detection
How we read each answer
Mentioned
We strip image markup and link URLs (keeping the link text), then look for your brand name, your domain, and your domain's root word (acme from acme.com, if it's at least 4 letters and not a dictionary word), as whole words and ignoring case. Simple variants count too: with or without spaces and hyphens, and without suffixes like “Inc.” or “Labs”. If your brand is a common English word, we require the exact capitalization, or your domain in the same answer.
Rank
We read the answer's structure: numbered and bulleted items, subheadings and table rows. Each item's name is its first bold phrase, else its link text, else the text before a dash or colon. Generic headings like “Summary” or “Pricing” are skipped. Your rank is the position of the first item that names you. If you're only mentioned in a sentence, we place you by where you first appear among the recommended names.
Cited
Your domain is cited when a source attached to the answer, or a link inside it, points to your domain or one of its subdomains. We ignore “www.”, query strings and fragments. A lookalike such as notacme.com doesn't count for acme.com.
Competitors
Competitors you enter count wherever they appear. Any other name counts once it's recommended in at least 2 answers across the scan, so one-off mentions don't crowd the list. You are never your own competitor.
Sentiment (heuristic)
A small word list scores the sentences around your mention, plus your own list entry: positive, neutral or negative. It's a rough signal, and the report labels it as one. It's empty when you aren't mentioned.
Known (brand prompt)
On the brand prompt, an engine “doesn't know you” if it never mentions you, or says things like “not familiar with”, “couldn't find”, “no information” or “may be a smaller company”. We also keep 1–2 sentences around your first mention (up to 320 characters) as a snippet.
No AI answer
Sometimes there's nothing to analyze, most often when Google shows no AI Overview for a search. Those answers are marked “no AI answer” and left out of every percentage and the score. Failed requests are retried; if they still fail, they're marked as errors and left out too.
4. Scoring
How the AI Visibility Score is calculated
Points for one answer
| In one buyer-prompt answer | Points |
|---|---|
| Your brand is mentioned | +50 |
| Ranked #1 in the answer's list | +30 |
| Ranked #2 or #3 | +20 |
| Ranked #4 or #5 | +10 |
| Ranked #6 or lower, or no list | +0 |
| Your domain is among the cited sources | +20 |
| Not mentioned, but your domain is cited | 20 |
| Neither mentioned nor cited | 0 |
| Maximum per answer | 100 |
Examples
- Mentioned at #1 and your site is cited100
- Mentioned at #2, not cited70
- Mentioned at #4 and cited80
- Mentioned in passing, no ranked list50
- Not mentioned, but your site is cited20
- Not mentioned, not cited0
Mentioned answers are capped at 100. Rank comes from the answer's list; a mention in prose is placed by order of appearance.
The overall score
AI Visibility Score = round( mean of answer scores over answered buyer prompts )
Worked example (5 answered buyer prompts)
- ChatGPT · P170
- Perplexity · P1100
- Claude · P20
- Gemini · P320
- AI Mode · P260
round((70 + 100 + 0 + 20 + 60) / 5) = round(50.0) = 50 → Visible
Score bands
- Invisible
- 0–19
- Emerging
- 20–44
- Visible
- 45–69
- Leading
- 70–100
The other numbers in your report
- AI Visibility Score
- The mean of the answer scores across answered buyer prompts, rounded to a whole number (0–100).
- Per-engine score
- The same mean, for one engine's answered buyer prompts.
- Mention rate
- Answered buyer-prompt answers that mention you, divided by all answered buyer-prompt answers.
- Citation rate
- Answered buyer-prompt answers that cite your domain, divided by all answered buyer-prompt answers.
- Average rank
- The mean of your rank in the answers where you were ranked.
- Share of voice
- Your mentions divided by your mentions plus competitor mentions, across answered buyer prompts. Each competitor counts at most once per answer.
Your report is built from whatever has finished, so the numbers update live while a scan runs and settle when the last answer is in.
Why it's built this way
Real answers, not a readiness score
Some AI visibility checkers grade your website's markup and estimate how “AI-ready” it is. That's useful hygiene, but it can't tell you whether an engine actually recommends you.
We ask the engines, with the questions your buyers ask, and show you every answer and source. If the number looks wrong, you can read the evidence and decide for yourself.
5. Limits
What this method can't tell you
Answers vary from run to run
Ask the same engine the same question twice and the list can change. One scan is a sample, so re-check monthly and compare trends rather than single numbers.
What you see logged in can differ
ChatGPT and Gemini answers are captured from the live apps where possible; Claude and Perplexity answers come from their APIs. A logged-in user can get a different answer because of personalization, memory, location or model version.
Claude may answer from training data
Depending on our configuration, Claude answers without a live web search, so its answer reflects what the model learned in training and can lag behind recent changes.
AI Overviews don't show for every search
When Google shows no AI Overview, that answer is marked “no AI answer” and doesn't count for or against you.
A small sample by design
3 buyer prompts on 6 engines is a quick health check, not a market study. Your buyers ask many other questions; the tracker and audit cover more of them.
Detection is automated
Brand names that are common words, or close to another company's name, can cause a miss or a false match. That's why the full report shows every answer, so you can check our reading.
Recent answers may be reused
If the exact same prompt went to the same engine for the same country in the last 24 hours, we may reuse that answer instead of asking again. Reused answers are marked as cached.
English, five countries
Prompts are in English. You can pick United States, United Kingdom, Canada, Australia, India. Google surfaces and the app captures use that location; the APIs use it where they support it.
Demo data is always labeled
If a report shows a “Demo data” badge, its answers are generated examples rather than live AI answers. Real scans never carry that badge.
6. Privacy
Your data
What we send to AI engines
Only the prompts: your brand, domain, category, audience and competitors, plus the country. Never your email. They go through our data provider, DataForSEO, to the engines.
We don't crawl your website
The scan reads what AI engines say. It doesn't fetch, copy or analyze your site.
Email only when you choose
Your score and per-engine results are free to see. Your email unlocks the full report and lets us send you the link.
Unlisted reports
Reports live at an unguessable link and are hidden from search engines. Anyone you share the link with can open the report.
Retention and deletion
We store hashed IP addresses, never raw ones, and keep scans for 12 months. Ask us and we'll delete your scan and lead data sooner.
Methodology questions
Not always. ChatGPT and Gemini answers are captured from the live apps where possible, with web search on, and fall back to the official APIs if the capture fails. Claude and Perplexity come from their APIs. A logged-in user can see something different because of personalization, memory and location. That's why we show the raw answers and recommend re-checking over time.
AI engines don't return a fixed ranking. The same question can produce a different list minutes later, and engines update their models and sources. Treat a single scan as a sample and look at the trend across several.
The engine returned nothing to analyze. Most often Google simply didn't show an AI Overview for that search. Those cells are excluded from every percentage and from the score, so they never count against you.
No. We don't fetch your site at all. We ask AI engines buyer questions and analyze their answers. The score reflects what AI says about your category, not an estimate built from your site's markup.
Because it names you. Any engine will talk about a brand you ask about directly. The score only uses unbranded buyer prompts, which test whether AI brings you up on its own. The brand prompt is used to check whether engines know you and how they describe you.
Yes. Unlock the full report with your email to read every answer, with your mentions highlighted, the sources each engine cited and the competitors it named.
Prompts run in English. You can choose United States, United Kingdom, Canada, Australia, India. Google AI Overviews, AI Mode and the app captures use that location, and the APIs use it where supported.