How Seenby measures AI visibility
Last updated 3 November 2026
In short: Seenby sends your prompts, exactly as written, to each connected assistant’s official API with web search on, asks more than once, reads each answer for brand names and cited pages with fixed rules, and reports shares over many answers rather than single results.
1. Which assistants, and how they are asked
| Assistant | Official API used |
|---|---|
| ChatGPT | OpenAI Responses API with web search |
| Perplexity | Perplexity Sonar API |
| Gemini | Gemini API with Google Search grounding |
| Google AI Overviews | SerpApi (Google results with AI Overview) |
Each assistant is switched on by its own API connection. The app, the free check and every report show which assistants are connected; an assistant that is not connected is shown as such, never with placeholder answers. No system prompt mentions your brand: the prompt is sent as a buyer would type it. For Google AI Overviews, a brand’s market (for example Pakistan) sets the Google country.
2. Sampling
Answers vary between runs, so Seenby asks each prompt more than once where the plan allows and reports shares: Starter asks each prompt 2 times per run, weekly; Growth asks each prompt once per run, every 3 days; Agency asks each prompt once per run, weekly. Trends over several runs are more reliable than any one week. Every plan has a monthly cap on assistant calls, so costs stay fixed.
3. Mention detection
- Each brand and competitor is matched by its name, any aliases you add, and its domain (for example acme.com).
- Matching ignores case and requires whole words in any script: “Notion” does not match “notional”; Urdu and Arabic names work.
- Possessives count (“Acme’s pricing”), and Markdown formatting such as bold is ignored.
- Position is the order in which tracked brands first appear in the answer: 1 means named first.
4. Sentiment
The sentences around a brand’s first mentions are scored with a fixed list of positive words (such as recommended, reliable, affordable) and negative ones (such as expensive, limited, complaints). The result is positive, neutral or negative. It is a simple, transparent heuristic, meant to flag answers worth reading, not a precise measure.
5. Citations
Cited pages come from the API’s own source list (URL citations, search results or grounding sources). Seenby groups them by domain and by kind: your own site, competitors’ sites, review sites, community threads, “best of” and comparison articles, video, news and reference. Those groups drive the recommendations.
6. The numbers
- Mention rate = answers that name you ÷ all answers. Shown as “named in 7 of 20 answers”.
- Share of voice = your mentions ÷ (your mentions + all tracked competitors’ mentions), over the same answers.
- Cited = answers that cite a page on your own domain.
- Google searches with no AI Overview are recorded as “no AI answer”, not as a miss.
More in the guide to AI share of voice.
7. Limits we are honest about
- API answers are close to, but not identical with, what a signed-in person sees in an app.
- Results depend on the prompts and competitors you choose; keep them stable to compare over time.
- Sentiment is a word-list heuristic.
- Seenby measures and recommends; it cannot promise that any change will get you mentioned.
Questions
- Does Seenby scrape ChatGPT or other assistants?
- No. It calls each provider’s official API. Google AI Overviews come from a search-data API that returns Google results.
- Are API answers the same as what people see in the apps?
- Close, not identical. Seenby uses web-search-enabled API modes where they exist, which is the nearest official equivalent, but consumer apps can add personalisation, location and memory.
- How is share of voice calculated?
- Your mentions divided by your mentions plus all tracked competitors’ mentions, over the same answers. Mention rate is answers naming you divided by all answers.