01The engines and how we ask them
Paid plans ask ChatGPT, Claude, Perplexity and DeepSeek. The free scan asks Claude, Perplexity and DeepSeek.
We ask through each provider's API with web search switched on, the way a new user would ask: one question, no chat history, no account, and a plain instruction to answer helpfully, search the web when current information helps, and cite sources. ChatGPT and Claude search with their own provider's web search. Perplexity searches on every request.
- One run per question per engine on the free scan and the Monitor plan.
- Repeat runs on bigger plans: Grow asks every question twice per check, Agency three times. Headline numbers (visibility, share of voice, sentiment) use every run, so one unusual answer moves them less. Screens that list single answers show the first run.
- Every week paid plans re-ask the same questions, so each check can be compared with the last.
- Failed answers (a timeout or a provider error) are retried once. Answers that still fail are left out of every number.
02How the questions are picked
We read your homepage and build a profile: what you sell, who buys it, where you operate and who you compete with. From that profile a model drafts candidate buyer questions of three kinds: common questions most buyers in your category ask, niche questions about your specialty, and a question that names your brand.
Then we check the candidates against Google autocomplete. We send short search terms for your category to Google's public suggestion service and rank the candidates by how closely they match what people actually type. The free scan keeps 5: 3 common, 1 niche and 1 brand question. If autocomplete doesn't respond, we keep the model's own questions.
On a paid plan you can track 50 (Monitor) or 100 (Grow) questions, add your own and remove any. Sites in Spanish are asked, and answered, in Spanish.
03How we read each answer
A helper model reads every answer and records the brands it names, in the order they first appear; whether it names you (your brand name, its known variants or your domain); your position; how it talks about you; and every link it cites. Every number below is built from these readings.
04AI visibility
We work it out overall, per engine and per question. Example: 5 questions asked to 3 engines is 15 answers. If 3 of them name you, your visibility is 20%.
Each question also gets a status. Won: at least one engine names you. Gap: no engine names you, but competitors are named. None: no brand is named at all.
06Position
Your position in an answer is your place among the brands it names, counted in order of first appearance. Named first is position 1. Answers that don't name you have no position.
Average position (on Competitors) is the mean over the answers that name each brand. Best position (on Prompts) is the best place you reached for that question.
07Sentiment score
| How the answer talks about you | Points |
|---|---|
| Positive: recommends or praises you | 85 |
| Neutral: mentions you without a view | 55 |
| Negative: warns against you or reports complaints | 15 |
The helper model labels the tone of each answer that names you. We turn each label into points and average them over the answers that name you. Answers that don't name you don't count, and if no answer names you there is no score. Example: 3 positive answers and 1 neutral one score (85 + 85 + 85 + 55) ÷ 4 = 78.
08Citations and source types
Every link an engine returns with its answer is one citation. A link cited in two answers counts twice. A domain's share is its citations ÷ all citations. Your citations are links to your own domain or its subdomains.
| Source type | What it means |
|---|---|
| You | Your own site and its subdomains. |
| Competitor | The site of a competitor in your profile, or of another brand the engines name. |
| Directory | Review and listing sites, such as Clutch, G2, Trustpilot, Capterra, Yelp and Tripadvisor. |
| UGC | Communities and social sites, such as Reddit, Quora, YouTube, LinkedIn, Medium and TikTok. |
| Editorial | Publishers, such as Forbes, The Verge, GQ and Wikipedia, plus blog, news and guide pages on other sites. |
| Corporate | Any other company site. |
Each cited page is also tagged by format from its address: homepage, article, listicle, product page, profile, forum thread or video.
09Site readiness checks
We fetch your homepage, robots.txt, llms.txt and sitemap the way a crawler does, without running JavaScript, and run these checks:
| Check | Passes when |
|---|---|
| AI crawlers allowed | robots.txt doesn't block GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot or Google-Extended from the whole site. |
| llms.txt published | Your site serves a plain-text /llms.txt. |
| Structured data (JSON-LD) | Your homepage has JSON-LD structured data. |
| Readable without JavaScript | Your homepage text is in the raw HTML (more than about 600 characters). |
| Meta description | Your homepage has a meta description. |
| XML sitemap | /sitemap.xml exists, or robots.txt lists a sitemap. |
| Clear page heading (H1) | Your homepage has an H1 heading. |
| Served over HTTPS | Your site loads over HTTPS. |
| Google Preferred Sources button | Your homepage links Google's "Add as a preferred source" button for your domain. |
| Google Business Profile claimed | Google shows a Business Profile for your brand and it is claimed by its owner. Only scored when we could look it up. |
Site Audit also lists 11 AI and search crawlers and whether your robots.txt lets each one in.
10Demand and difficulty
Demand (0 to 5) estimates how often people search for a question. We take its first two key words, ask Google autocomplete for suggestions and count the suggestions that contain both words. Your questions are then ranked against each other from 1 to 5. 0 means Google suggested nothing. Demand compares your questions with each other: it is not a search volume.
Difficulty (1 to 5) counts how many other brands the engines name for a question. More brands in the answers means more competition for a spot.
| Other brands named for the question | Difficulty |
|---|---|
| 0 to 5 | 1 |
| 6 to 15 | 2 |
| 16 to 22 | 3 |
| 23 to 30 | 4 |
| More than 30 | 5 |
11Google AI Overviews and AI Mode
When the Google data feed is connected, we also search Google for each tracked question in your market and record whether an AI Overview appears, whether it names you, which sites it cites and whether your site is one of them. We do the same for Google AI Mode where Google offers it.
These results come from Google's live results pages, through a search data provider, not from the AI APIs above. They are saved with the check they belong to. Until the feed is connected, the Google AI screen shows your questions with empty cells and no estimates.
12What the numbers can't tell you
- API answers can differ from the apps. A signed-in person in ChatGPT, Claude or Perplexity gets answers shaped by memory, personal settings, location and sometimes a different model or extra features such as shopping cards. We ask with no history and no account, so we measure the answer a new user would get, as closely as an API allows.
- Answers vary from run to run. The same question can name different brands an hour later. That is why bigger plans repeat runs and every plan tracks trends over weeks. One check on its own is a snapshot.
- Small samples move in steps. With 30 answers, one answer is worth about 3 points of visibility.
- The reader can slip. The helper model can misread an answer, for example missing an unusual spelling of your name. You can open every answer in the app and check it.
- Citations are what the API returns. The consumer apps sometimes show more or fewer sources.
- Demand is relative. It is based on Google suggestions, not on search volume.
Questions about the method? Write to support@tryoptia.com.
Run a free scan