The fourteen checks
AI visibility
What ChatGPT and Gemini say about you when someone asks, which sources they lean on to say it, and how often you are mentioned in AI answers at all.
Check 12 of 14.
Parked in the alpha
This check does not run or bill in alpha reports. What follows describes it as built.
What happens
How often AI assistants mention you, which sources they cite when they do, what the ChatGPT app itself shows a person, and how that has moved month by month.
Calls it makes
DataForSEO POST ai_optimization/llm_mentions/target_metrics/live — mention counts and cited domainsPOST ai_optimization/llm_mentions/search/live — the questions, with the answersPOST serp/google/organic/live/advanced then POST serp/ai_summary — what the results page says about youPOST ai_optimization/{chat_gpt,gemini}/llm_responses/live — the model API's own answer, capped at 200 output tokensDEEPER SETTINGS ONLY — POST ai_optimization/chat_gpt/llm_scraper/live/advanced: the real chatgpt.com interface, with its sources, its search results and its ads ($0.004 measured)DEEPER SETTINGS ONLY — POST ai_optimization/llm_mentions/historical/live: month-by-month mentions bought from the provider's archive ($0.101 measured)DEEPEST SETTING ONLY — POST ai_optimization/llm_mentions/timeseries_new_lost/live: exactly what was gained and lost per month ($0.104 measured)DEEPER SETTINGS ONLY — the AI Gateway: six model families asked the same brand questions three times each, priced from each response's own token usageDEEPER SETTINGS ONLY — POST serp/google/ai_mode/live/advanced: DataForSEO's own read of Google's AI Mode for your category, the cheap bulk twin to the SerpApi lane's ai_mode unit ($0.004 measured — see the SerpApi provider entry for the paid alternative)DEEPEST SETTING + OPT-IN FLAG ONLY — the same grid, allowed to search: the AI Gateway again, this time only through routes that can look things up, priced from each response's own real cost ($0.005–$0.016/call measured — see the SerpApi/AI-visibility provider docs)
The facts a reader needs
| Language model | The grid is the third one. Six families answer the same questions through one key, three runs apiece, and every number carries how many runs it came from. Those calls are stateless and logged out and cannot search — the panel says 'from training, no search' because that is what they are. On a free-tier Gateway account each model rate-limits on a burst, so the grid PACES itself — bursting until a model 429s, then slowing only that model — and the panel says how many of the six answered inside the pull's own time budget rather than presenting a partial sweep as a complete one. |
| Cost | $0.22–$0.23 per pull at the standard setting, about $0.34 with the app surface, the trend and DataForSEO's own AI Mode read |
| Depends on | None. |
| When it is absent | WE STOPPED BUYING ONE THING. top_mentioned_brands cost eleven cents a pull and asked which brands are mentioned most for a topic that fell back to your own name — the question, not the answer. Reports produced before 2026-08-18 still show it; new ones do not buy it. MENTIONS ARE MATCHED ON WORD BOUNDARIES, never substrings: 'Ford' is not a mention inside 'afford'. SHARE OF VOICE IS A REAL PERCENTAGE — your namings divided by every brand naming in the same answers, with the denominator printed beside it, over a competitor set that films, characters and common English words are filtered out of before the arithmetic runs. THE CONTENT-GAP TABLE is that same word-boundary matching, per (model × question) cell rather than aggregated — 'the prompts where rivals appeared and you did not', capped so a prompt nobody named anybody in never counts as a missed opportunity. THE LOCATION DIAL IS COUNTRY-LEVEL, and it is the provider's limit rather than ours: their own location list holds 213 countries and two regions and no cities at all, so your city travels inside the question and the country travels in the parameter, and every panel says which was which. The archive also lags the present by a month or two, so a trailing zero on the trend line means 'not crawled yet' and is drawn as such rather than as a collapse. A MODEL THE GRID NEVER GOT TO ANSWER, because the pull's pacing budget ran out, is reported as rate_limited_budget or not_attempted_budget rather than silently dropped — see src/local/gateway-pacer.mjs. Adding paid Gateway credits removes the free-tier limit and lets the grid burst again automatically, no setting to change. THE GROUNDED LANE IS TIER 3 AND OPT-IN, on top of the tier — it roughly triples the ungrounded grid's spend, so a report does not buy it by default even at the deepest setting. Only Perplexity is genuinely grounded through this route today: verified live 2026-08-19, every other family's own 'web-search' tool is listed on the Gateway's model catalog but rejected by its OpenAI-compatible endpoint with an 'invalid tools.0.type' error, so those five families report not_supported_via_gateway — a named, honest absence, never a faked search dressed up in prompt text. Perplexity's location and its citations both travel differently here than its own direct API: the Gateway strips its native location parameter and its structured source list alike, so the city travels inside the question and sources are extracted from a line the model is asked to append, the same bracket-URL pattern the AI Overview read above already uses. |
Who supplies it
- DataForSEO — Search, SEO, reviews and backlinks. DataForSEO supplies the search data for paid reports and the beta weekly search check; its AI lanes stay parked, and the free read never calls it.
- Vercel AI Gateway — The door every model call goes through when AI_GATEWAY_API_KEY is set — the voice read, the confirm-step adjudicator, the quality review, the confirm-step suggestions and the AI-visibility grid. Unset it and the first four fall back to calling OpenAI directly, unchanged; the grid needs the Gateway and is parked in the alpha.
Voice
How you sound, written as claims — and every claim quotes the sentence of yours that proves it. A claim with no quote behind it is dropped, not softened.
Reviews
Your Google Business ratings — multi-location aware, so a chain is not collapsed into one listing — and your Trustpilot score with the reviews behind it.