Web research
answers that must cite sources.
Researches a claim on the web with a model that cites its sources, through Anthropic (Messages API with web search) or Perplexity (Agent API with web search). A claim no source is cited for is shown as No reliable source found.
01What it does
For each claim the language model found (at most three), the server makes one research call to your provider, in parallel with the fact-check search. Only the provider's structured citations count, at most five per claim; the model's own prose never decides anything:
| Cited sources | Finding |
|---|---|
| A cited web source supports the claim | OK |
| A cited web source contradicts it | Caution |
| Web research found no clear answer | Caution |
The truth answer per claim: "Supported by sources" needs support cited on two or more different sites (or a fact-checker's top rating); contradiction on two or more sites makes it "Disputed"; a single citation, citations from one site, or none at all stay "No reliable source found", never "true". Sites that appear in the checked content itself do not count as sources for it. Over several claims the worst answer wins.
An answer that carried no web search result at all is listed as "Web research: returned nothing", not as a finding. A busy or slow provider is "busy"; research may use the rest of the check's time budget.
02Terms of each provider
- Anthropic: the Commercial Terms let the customer own its outputs, and the web search documentation asks that citations are shown with the answer. Veritap shows every citation with its link, in the app and on shared verdict pages.
- Perplexity: the API Terms (section 2.1) allow displaying output "solely within the Customer Applications". Veritap therefore shows Perplexity research only inside the app: those findings are marked not shareable and never appear on a shared verdict page.
- Gemini is not offered: its grounding terms forbid storing grounded results, which Veritap needs for results, the locker and shared pages (see Outside services).
03Get a key
- Anthropic: create an account in the Claude Developer Platform, add billing, and create an API key under API keys. Pick a model id that supports the web search tool.
- Perplexity: create an account, add billing, and create an API key in the API settings. Pick a model id the Agent API offers with web search.
04Switch it on
- Console, Cited web research: turn on Use cited web research.
- Choose the Provider (owner only; changing it clears the saved key).
- Paste the API key and enter the Model id exactly as your provider names it. The model must support web search.
- Leave API base URL empty to use the provider's own address. A custom address must be https on a public address; it is checked when you save and before every call, and changing it clears the key.
- Press Test connection. It researches one fixed, neutral claim (one web search) and stops after 20 seconds.
Claims are found by the language model, so web research also needs the Language model.
05Cost
Billed to your Anthropic or Perplexity account. Anthropic charges $10 per 1,000 web searches plus tokens (pricing, read 12 November 2026); Perplexity publishes its prices at docs.perplexity.ai. One claim check makes up to three research calls, one per claim, and a call may run more than one search. Check the provider's pricing page before you launch.