DataForSEO Brain

AI Optimization API

5 min read

AI Optimization API

A data layer for querying LLMs, scraping their consumer answers, and measuring AI search volume. Sits under DataForSEO Brain → Concepts.

Overview

The AI Optimization API is DataForSEO’s interface to the generative-search era. It is raw infrastructure (no dashboard) split into four endpoint groups: LLM Responses (call models via API), LLM Scraper (scrape the consumer ChatGPT/Gemini web UI), AI Keyword Data (AI search volume), and LLM Mentions (brand/citation visibility, covered separately in cap-llm-mentions-visibility). It lets teams test what each model says about a brand, collect cited sources, and size AI demand programmatically. Everything is pay-as-you-go with sandbox testing available; LLM Mentions joined that model on 2026-07-01 with no subscription or activation step.

What it covers

  • LLM Responses (/v3/ai_optimization/{provider}/llm_responses/): structured answers from ChatGPT, Claude, Gemini, and Perplexity through one interface. ChatGPT/Claude/Gemini support Live and Standard; Perplexity is Live-only (task_post_supported: false). Each provider has a /models/ endpoint that is the source of truth for accepted model_name strings: the live /v3/ai_optimization/claude/llm_responses/models endpoint currently accepts claude-sonnet-4-6, claude-opus-4-8, claude-sonnet-4-5, and claude-opus-4-5, among others. The static docs page lists older models and lags Anthropic’s release cadence, so always re-poll the /models/ endpoint before pinning a version. Live LLM Responses calls are pay-as-you-go accessible with no subscription (verified 2026-06-26; fixtures captured at .raw/sources/dataforseo-research/ai-optimization/fixtures/llm-responses-*.json).
  • LLM Scraper (/v3/ai_optimization/{provider}/llm_scraper/): scrapes the live ChatGPT and Gemini web interfaces for a keyword, returning the markdown answer, search_results[], sources[], typed items[] (text, tables, images, products, local businesses), and brand_entities[]. Advanced and HTML output. Recent change (2026-05-28): the ChatGPT LLM Scraper now returns ad results. It surfaces the sponsored / ad placements that appear inside ChatGPT responses, so the scraped items can now include paid results alongside the organic answer - relevant for tracking how ads show up in AI answers and for GEO competitive monitoring.
  • AI Keyword Data (/v3/ai_optimization/ai_keyword_data/keywords_search_volume/live): ai_search_volume and 12-month ai_monthly_searches, calculated from People Also Ask SERP data. Live-only, up to 1,000 keywords.
  • LLM Mentions (/v3/ai_optimization/llm_mentions/): the brand and citation visibility tracker, split out into its own note cap-llm-mentions-visibility. It is pay-as-you-go since 2026-07-01, with no subscription or activation step; /v3/ai_optimization/llm_mentions/search/live returned 20000 Ok in the 2026-07-08 live fixture at .raw/sources/dataforseo-research/ai-optimization/fixtures/llm-mentions-search.json.
  • Reference endpoints: each provider has a /models/ endpoint to enumerate available model versions, and AI Keyword Data has /locations_and_languages/ for valid targeting.

Key parameters / inputs

fieldnotes
user_promptLLM Responses; required; up to 500 chars
model_namerequired; basic name resolves to latest version
keywordLLM Scraper; up to 2000 chars
keywordsAI Keyword Data; max 1000, up to 250 chars each
max_output_tokens / temperature / top_pgeneration controls (temperature and top_p cannot combine); max_output_tokens ranges 16-2048 for non-reasoning models and 1024-4096 for reasoning models
web_search / force_web_searchenable or force the model to cite current web info (force_web_search is ChatGPT/Claude only)
web_search_country_iso_code / web_search_citygeo-target the web grounding (e.g. “FR”, “Paris”); ChatGPT and Claude only
use_reasoningClaude and Gemini reasoning mode (min ~1024 tokens)
message_chain / system_messageup to 10 prior turns; role/tone control
location_* / language_*LLM Scraper and AI Keyword Data targeting

Response / what you get back

LLM Responses returns model_name, input_tokens, output_tokens, reasoning_tokens, web_search, money_spent (the third-party model’s token charge, passed through), items[] (reasoning and message objects, each with sections[] of type/text and optional annotations[] of web sources with title and url), and fan_out_queries[]. LLM Scraper returns markdown, sources[], search_results[], typed items (e.g. gemini_text, gemini_table, gemini_images, gemini_source), and brand_entities[]. AI Keyword Data returns ai_search_volume and ai_monthly_searches[].

Cost & method notes

  • Pricing combines DataForSEO’s per-task cost with the model’s pass-through money_spent. LLM Responses Live is 0.0002 plus a 0.0006 Standard figure. LLM Scraper (ChatGPT and Gemini only) is 0.0024 Priority, 0.01/task plus $0.0001/keyword.
  • In the cost log, AI Keyword Data search_volume ran $0.0101. See cap-queue-priority-cost-model and cap-task-vs-live-execution.

When to use / how it fits

Use LLM Responses to A/B what models say about your brand vs competitors, LLM Scraper as the budget option for real consumer answers with brand detection, and AI Keyword Data to find rising AI-phrased queries. This API powers play-ai-visibility-tracking and underpins cap-geo-ai-search-optimization. The models behind it are profiled in ent-llm-model-providers and the answer surfaces in plat-ai-assistants. Route by job in dec-which-api-for-which-job.

Gotchas / limits

  • Execution time is up to 120 seconds per task; 30 concurrent Live tasks per account per platform; 2000 calls/min; one task per Live call.
  • annotations can return empty even when web_search is true; some ChatGPT models (o3-mini, o1, o1-pro) do not support web_search.
  • Perplexity Sonar models use web_search by default and are Live-only.
  • money_spent is variable and stacks on top of the base task cost, so budget for both.
  • LLM Responses temperature and top_p cannot be used together, and Claude cannot combine force_web_search with use_reasoning.
  • Reasoning models can produce output that exceeds max_output_tokens when web search or reasoning is enabled.

Sources

Graph View

Backlinks