AI visibility tracking
Track fixed AI prompts, answers, mentions, citations, model changes and exact provider costs without inventing a universal visibility score.
Available now
Record fixed AI prompt observations locally
Run a small fixed set of prompts against explicit current models in ChatGPT, Claude, Gemini, or Perplexity. Each observation keeps the answer, citations, model the provider actually ran, token use, cache state, provider task id, and exact returned cost.
The report stores limited history on this computer and compares only compatible observations. It does not turn a handful of generated answers into a universal visibility score or assistant ranking.
What the report records
The evidence stays attached to every summary. Read the answer, model, settings, what each source returned, and collection time before interpreting a mention or change.
| Measurement | What it tells you |
|---|---|
| Answers | The returned answer for each exact prompt, surface, chosen model, model the provider actually ran, settings, and time. |
| Named targets | Whether supplied brand aliases appeared in that answer sample, kept separate from citations. |
| Citations | Which pages and domains appeared in returned citation metadata, including citations that do not name the target. |
| Prompt set coverage | How many completed observations in the fixed set included the target or one of its domains. |
| Supporting searches | Extra searches the provider reports using to form an answer. The structured result calls these fan-out queries. |
| Like-for-like change | Which answers, citation domains, or named targets changed between complete observations with the same configuration. |
Cached observations, model changes, partial answers, and truncated answers stay non-comparable. Missing, unavailable, and not observed remain different states.
Use the right evidence for the question
Live answers, indexed mention research, technical eligibility, and recorded visits answer different questions. Keep them separate until a decision needs them together.
Record current answers, mentions, citations, supporting searches, like-for-like changes, and exact costs for a small fixed set of prompts.
Research provider-indexed mentions, cited domains, and limited question samples for one surface and market.
Check crawl access, indexability, snippet controls, page structure, and optional agent resources.
Check the technical controls Google documents for AI search eligibility.
Find visits from known AI assistants that reached the site and were recorded by Google Analytics.
Turn real search queries into a stable prompt set for an external tracking provider.
A crawl restriction can affect eligibility. A live answer is one generated sample. A provider-indexed mention is limited by that dataset. A Google Analytics referral proves that one recorded visit reached the site. None can replace the others.
Run a small fixed set of prompts
The report accepts no more than five prompts and four models, for a maximum of 20 paid live requests. Start smaller and add only prompts that can change a decision.
- 01A local DataForSEO connection with spend and request limits set.
- 02One to five fixed prompts, with stable ids and optional groups.
- 03One to four exact current model names from ChatGPT, Claude, Gemini, or Perplexity.
- 04Your target aliases and domains, plus no more than five named competitors.
- 05An optional Search Console property when Search Console query context would help.
Provider and model rules
DataForSEO supplies live AI responses now. The report contract is provider neutral, so another provider can implement the same capability later without changing how the report is read.
Model names must match the provider's free current catalog. An invalid or stale name fails before a paid request and lists current choices. The base-price estimate stays separate from the exact task cost returned after completion.
What AI visibility data still cannot prove
AI answers can vary on another request. A fixed prompt set gives you a repeatable sample across the prompts, models, market labels, settings, and dates you choose. A missing mention only means the returned answer did not contain a supplied alias or domain. It does not prove the target never appeared elsewhere.
Citations need review. A returned URL can be irrelevant or attached to the wrong claim. Supporting-search themes and Search Console word matches are research leads, not proof of demand or permission to build a programmatic template.