How to see OpenAI search queries
To see OpenAI search queries, run a topic through the Responses API with web search enabled and inspect the returned search-action fields. The free tool on this site shows the query strings and sources that response exposes.
Use this workflow to investigate a content topic or inspect a search result. It observes a new API run, rather than accessing your ChatGPT conversation or private search history.
Run an observable OpenAI searchChoose OpenAI and inspect only the query strings and sources present in that API response.Show OpenAI queries- Open the tool with OpenAI selected
Enter one topic of up to 60 characters. Choose language and country to match the audience you are researching.
- Run the search and read the counts
Complete the security check and start the run. Read search actions, exposed queries and cited sources separately; one action may contain several queries.
- Take selected queries into your brief
Copy or export the useful query strings. Check source links at the scope shown and validate human demand before planning a page.
Read one dated observation completely
Start with the published best-seo-tools-us example from 27 August 2026 at 13:13:23.681 UTC. Its input is best SEO tools, language en, country US. The example contains four normalized strings and one search action. Four strings therefore do not establish four separate searches.
providerResponseStatus is incomplete. The search action returned material while the final answer reached the token cap. Describe it as observed search material under that protocol, not a complete final answer. Recorded usage is 12,905 input and 500 output tokens; it does not measure human demand.
Continue with review the source evidence.
The published file is a normalized observation dataset, not raw Responses JSON. sourceDomains contains domain names without a per-query join. It cannot reconstruct organic rank or a supported source for each string. Present date, status and file type together when explaining this example.
What this specific example establishes
An actual previously published OpenAI observation; no new API call or synthetic raw response.
| Field | Value | Meaning |
|---|---|---|
| observedAt | 2026-08-27T13:13:23.681Z | Time of this example |
| queries.length / usage.searchActions | 4 / 1 | Four strings in one search action |
| providerResponseStatus | incomplete | Partial status retained |
| sourceDomains | Example-level domain set | No individual query join |
Separate raw response from website result
In raw responses, exposed strings appear in query or queries of a web_search_call search action. Answer-text URL annotations have different scope from action sources. The current API reference supplies these field paths; the published normalized example does not replace that structure.
The website returns its own full run with keyword, providerId, modelId, generatedAt, queries and searchActions. toolVersion and methodVersion identify the implementation. Its wrapper trims, normalizes, deduplicates and caps the list at eight strings. An export is therefore not a lossless raw-response archive.
For an existing run, choose its appropriate export and inspect metadata. For your own parser code, use the separate developer tutorial. That offline example preserves exact strings and counts only search actions as searches; it deliberately differs from website normalization.
Continue with build and test an OpenAI parser.
Zero, partial output and errors are different results
A completed run without exposed search strings can be a valid empty observation. Its supported statement is zero visible strings in this run, not that the provider never searches or the topic has no demand.
Timeouts, HTTP failures and malformed JSON are not successful zero observations. For incomplete output with exposed actions, retain strings and partial status together. A failed attempt has an unknown count. This prevents request failures from entering comparisons as false zero values.
The distinction matters when reading the current site: it supplies fallback values for some absent status fields and treats OpenAI incomplete output without a visible web call as an error. A stricter consumer must preserve raw status separately. The methodology covers bounded API evidence, not private ChatGPT search histories.
Review the published methodology.
Common questions
These short answers cover recurring questions from tool use and source review. They are not claimed search-volume data.
- Does ChatGPT Search rewrite a prompt into targeted searches?
- OpenAI says ChatGPT Search may rewrite a question into one or more targeted queries for search providers and may issue more specific follow-up searches. That consumer-product description does not give this site access to a private or complete ChatGPT search history.
- Can the OpenAI API expose the actual search queries from one run?
- It can expose a query or list of queries in documented web-search action fields. The safe claim is limited to the strings present in that named API response; missing strings remain unknown.
- Why can one search action contain four queries?
- A provider may bundle several query strings into one search action. The tool preserves that distinction: search actions count requests, while queries count the visible strings inside them.
- Why are the input tokens much higher than my keyword?
- Provider usage can include tool instructions, retrieved context and other API-side input, not just the short keyword. The displayed token count comes from the run metadata and is not a character count of your entry.
- Can zero visible queries be a valid result?
- Yes. A completed run can expose no search strings, including when sources appear without a query field. A failed or incomplete attempt must retain that state; it is not a successful zero observation.
Sources used
These sources support the functions and limits described here. They do not prove claims beyond that scope.
- Original runnable offline parser and edge-case test kitOpen source
Original educational parser code and 26 offline checks. Synthetic response fixtures are not provider output; exact strings, status states, source scope and retention exclusions are tested. No paid calls or new actual grounding data. The teaching helper explicitly differs from the live normalized wrapper.
- Current website export contracts and reference validationOpen source
Draft 2020-12 reference for the current native/modelled full runs, query selections and local comparisons. Schema checked against actual service/export functions with synthetic provider stubs and semantic negative tests. No root schemaVersion is invented for full runs. Legacy planner and handoff formats remain separate.
- OpenAI Responses: web-search action fieldsOpen source
Supports exposed action query fields, distinct search/open/find action types and URL annotations. The documentation does not provide private consumer traces or guarantee every run exposes strings.
- OpenAI Responses API referenceOpen source
Supports one stateless Responses API call, store false, max output tokens and strict JSON Schema output.
- OpenAI web search tool guideOpen source
Documents web_search_call output and search actions that usually, but not always, include the query or queries searched. Search actions, exposed strings and cited sources therefore need separate interpretation. It does not expose chain of thought or guarantee that every route returns query strings. Rechecked for the original 2026-10-10 zero-query decision table; no new provider observation is published.
- Searching the web with ChatGPTOpen source
OpenAI explains that ChatGPT Search can rewrite a question into one or more targeted queries and can issue more specific follow-up searches; this description does not expose a private ChatGPT search history to ai-fanout.com.
- GPT-5.6 Luna model pricingOpen source
Lists USD 0.20 per million input tokens and USD 1.20 per million output tokens used by the dated per-run estimate.
- OpenAI web-search pricingOpen source
Lists the standard web_search tool at USD 10 per 1,000 calls plus search-content tokens at model rates.
- Direct OpenAI and Gemini native-fanout implementationOwner record
One direct provider request, strict provider allowlist, provider-exposed query/source parsing, honest zero-query output, CAPTCHA-first shared atomic quota and budget accounting, no raw server archive and separate public flag. Bounded live production fixtures passed for OpenAI and Gemini on 2026-08-26.