Compare LLM Search Behavior Through One API.

Copy the complete instructions or download them as CLAUDE.md, then place the file at the root of your project. The same instructions also work when pasted into Claude, Codex, Cursor, or another coding assistant.

Use the API when the user wants to compare how AI answer engines research and answer the same question. For a meaningful comparison, select at least two providers unless the user explicitly asks to inspect one model.

• Compare the search queries generated by different providers and models.
• Find which domains and pages ChatGPT, Claude, Gemini, or Perplexity cite.
• Inspect extracted page evidence behind those citations.
• Audit whether a company, product, domain, or page appears in AI answers.
• Compare answer framing, recommendations, omissions, and provider failures.

Provider agreement is cross-model consensus, not proof that a claim is true. Use the returned source evidence to evaluate it.

Ask the user for an LLM Sleuth API key before making a live request. Keys start with asc_. Never expose a key in browser code, source control, logs, or assistant output.

Authorization: Bearer asc_your_api_key

The x-api-key header is also accepted. Create or rotate keys at llmsleuth.com/dashboard.

1. List available models

Fetch the catalog before every comparison. It is the source of truth for current model IDs, provider IDs, defaults, availability, and credit costs.

curl https://llmsleuth.com/api/v1/models \
  -H "Authorization: Bearer asc_your_api_key"

{
  "object": "list",
  "models": [
    {
      "id": "gpt-5.6-sol",
      "provider": "openai",
      "name": "GPT-5.6 Sol",
      "credit_cost": 2,
      "is_default": true
    }
  ]
}

2. Choose explicit model runs

Use runsby default and match every model to the provider returned by the catalog. Add the selected models' credit costs so the user can approve an expensive comparison before it runs.

3. Run the comparison

curl https://llmsleuth.com/api/v1/query \
  -H "Authorization: Bearer asc_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "What is the best lead scraping tool?",
    "runs": [
      { "provider": "openai", "model": "gpt-5.6-sol" },
      { "provider": "anthropic", "model": "claude-sonnet-5" },
      { "provider": "gemini", "model": "gemini-3.5-flash" }
    ]
  }'

Send either runs or providers, never both. The request is synchronous and can take several minutes, so use a client timeout of at least seven minutes.

Do not automatically retry an ambiguous timeout. A retry can create a second charged execution; check dashboard history or ask the user first.

A successful HTTP response can still contain an individual provider failure. Inspect every results[].ok value and report partial failures alongside the successful runs.

{
  "execution_id": "7f79c814-1112-4a94-9bb4-719d5ccf5e98",
  "query": "What is the best lead scraping tool?",
  "runs": [
    { "provider": "openai", "model": "gpt-5.6-sol" },
    { "provider": "anthropic", "model": "claude-sonnet-5" },
    { "provider": "gemini", "model": "gemini-3.5-flash" }
  ],
  "providers": ["openai", "anthropic", "gemini"],
  "credits_used": 4,
  "results": [
    {
      "provider": "openai",
      "label": "ChatGPT / OpenAI",
      "ok": true,
      "model": "gpt-5.6-sol",
      "generatedQueries": ["best lead scraping tools 2026"],
      "sources": [{ "url": "https://example.com/source" }],
      "sourcePreviews": [
        {
          "url": "https://example.com/source",
          "fetchStatus": 200,
          "headings": [{ "level": 1, "text": "Best Lead Tools" }]
        }
      ],
      "answer": "Markdown answer text..."
    }
  ]
}

Keep every finding attributable to its provider. Unless the user asks for another format, organize the result around:

1. Search strategy: how each provider framed the research.
2. Source overlap: domains or exact URLs cited by multiple models.
3. Unique sources: sources used by only one model.
4. Evidence quality: primary evidence versus thin, blocked, redirected, or missing pages.
5. Answer differences: recommendations, claims, caveats, and omissions.
6. Visibility: whether the target appeared in each answer or source list.
7. Failures and limits: provider errors and unavailable query or preview data.

Legacy default-model request

Use providers only when current default models are acceptable and exact model selection does not matter.

curl https://llmsleuth.com/api/v1/query \
  -H "Authorization: Bearer asc_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Which CRM tools are most often recommended to startups?",
    "providers": ["openai", "anthropic", "gemini", "perplexity"]
  }'

• Fetch GET /api/v1/models before choosing models because availability and costs can change.
• Selected model costs are charged before the comparison runs.
• Missing generated queries mean unavailable capture, not proof that the provider did not search.
• Only the first six sources are enriched as previews; the full source list remains in sources.
• A failed preview does not erase the citation. Preserve the URL and report the fetch limitation.
• A citation shows model behavior; it does not automatically validate the cited claim.
• Save the returned execution_id so the comparison can be found in history.
• Use https://llmsleuth.com for production integrations, never localhost or a preview domain.