DataForSEO Brain

Play: AI Visibility Tracking

4 min read

Play: AI Visibility Tracking

The GEO loop: measure brand mentions/citations inside AI answers with LLM Mentions, sample fresh model answers with LLM Responses, and roll it into a share-of-voice dashboard. Sits under DataForSEO Brain → Flows.

Overview

Generative Engine Optimization (GEO) tracks whether a brand is named, cited, or recommended inside AI answers rather than ranked in blue links. This flow has two engines: the LLM Mentions API (a pre-indexed mentions/citations database with its own AI search-volume index) for aggregated visibility, and LLM Responses (live model querying across ChatGPT, Claude, Gemini, Perplexity) for fresh prompt sampling. Together they give a repeatable AI-era rank tracker.

Trigger

A brand wants to know how it shows up in AI answers: a GEO program launch, a competitive share-of-voice study, or monitoring after a content push. Inputs are target entities (brand domain and/or keywords) and a prompt set of real buyer questions.

Endpoints used (in order)

  • POST /v3/ai_optimization/llm_mentions/search/live (raw mention records).
  • POST /v3/ai_optimization/llm_mentions/aggregated_metrics/live (consolidated metrics).
  • POST /v3/ai_optimization/llm_mentions/cross_aggregated_metrics/live (multi-brand comparison).
  • POST /v3/ai_optimization/llm_mentions/top_domains/live and /top_pages/live (most-cited sources).
  • POST /v3/ai_optimization/{chat_gpt|claude|gemini|perplexity}/llm_responses/live (live answer sampling).
  • POST /v3/ai_optimization/ai_keyword_data/keywords_search_volume/live (AI search volume).

Pipeline

  1. Define targets: build the target array (up to 10 entities) of domains (domain, search_filter, search_scope sources/search_results) and keywords (keyword, match_type, search_scope question/answer/brand_entities). Set platform (google for AI Overview, chat_gpt) plus location_code/language_code.
  2. Pull mentions with llm_mentions/search: each item returns the question, the answer (markdown), cited sources[] (title, url, domain, position), search_results[], brand_entities[], fan_out_queries[], and ai_search_volume with monthly_searches[].
  3. Aggregate visibility with aggregated_metrics (mentions + ai_search_volume grouped by location/language/platform/source domain) and compute share of voice = your mentions / total tracked-brand mentions.
  4. Benchmark competitors with cross_aggregated_metrics: supply 2-10 targets each with an aggregation_key label to get side-by-side group totals in one call.
  5. Find citation surfaces with top_domains/top_pages to see which domains and pages the models cite for your topic (often third-party/UGC, not your own site).
  6. Sample fresh answers with llm_responses: send the same prompt set across ChatGPT, Claude, Gemini, and Perplexity (user_prompt ⇐500 chars, choose model_name, set web_search: true to force current citations) and parse items[] message sections[] and annotations[] for mention presence and sentiment.
  7. Score and chart: per engine, record mention rate, citation rate, share of voice, position, and sentiment over time; alert on drops.

Cost & cadence

  • LLM Mentions is Live-only and pay-as-you-go, priced about 0.001/row, roughly $1.10 per 1,000 rows; turnaround about 2 seconds. It returned 40204 through 2026-06-26, then the subscription and activation step were removed on 2026-07-01. A 2026-07-08 re-probe of /v3/ai_optimization/llm_mentions/search/live returned 20000 Ok with real data; fixture: .raw/sources/dataforseo-research/ai-optimization/fixtures/llm-mentions-search.json.
  • LLM Responses bills DataForSEO’s per-task cost (about 0.0101/task (cost-log: $0.0101).
  • Cadence: mentions/aggregates weekly or monthly (the data has churn but stable patterns); live prompt sampling weekly across the model set.

Output

An AI visibility dashboard per engine: mention rate, citation rate, AI share of voice, sentiment, the most-cited source domains/pages, and the fan-out queries driving answers. Pairs with play-content-strategy-brief to act on citation gaps.

Pitfalls / limits

  • Coverage is narrower than the headline: ChatGPT mentions data is US (location_code 2840) and English only; Google AI Overview supports multiple locations/languages.
  • impressions is deprecated and returns null; do not build on it.
  • LLM Responses annotations[] can be empty even when web_search: true, and web_search is unsupported on some reasoning models (o1, o3-mini); validate per model.
  • Most brand validation is off-domain (third-party pages), so track earned/UGC sources, not just your own site.
  • Track each engine separately; cross-engine citation overlap is low, so a single blended number hides the real picture.

Decisions in play

  • dec-which-api-for-which-job: LLM Mentions for aggregated brand/citation visibility; LLM Responses for fresh, controllable prompt sampling across models.
  • dec-cost-control-strategy: LLM Mentions bills per request and per row, so design a fixed prompt/target set and cache; LLM Responses adds a pass-through money_spent token charge per model.
  • dec-dataforseo-vs-ahrefs-semrush-moz: DataForSEO is raw GEO infrastructure, not a finished SOV dashboard; you build the scoring and charts.

Sources

Graph View

Backlinks