Query fan-out › Data

Data

From QueryFanout.wiki, the technical reference and research archive for query fan-out

The raw records of the Query Fan-Out Observatory, published as versioned CSV and JSON under Creative Commons Attribution 4.0. Each row is one test of one prompt on one platform at one time, with the subqueries the platform displayed, the sources it retrieved and cited, and an evidence class. The method is on the Research page. Use the data freely; cite queryfanout.wiki and the version number.

query-fanout-tests.csv download
query-fanout-tests.json download
version 0.1.0
rows 13
published 2026-09-08
licence CC BY 4.0

Batch 1 summary

TESTED Eleven frozen prompts, one per prompt class, run on Perplexity (signed-in free account, Colorado, USA) on 8 September 2026 between 03:08 and 03:20 UTC, which was the evening of 7 September in Colorado, plus two repeat runs of the multifaceted prompt. Twelve rows are OBSERVED: the platform displayed its searches and they were captured as shown. One row (the follow-up turn) is recorded with fanout_observable false because the platform displayed no step panel. Google AI Mode rows are absent from this batch for the reasons on the Measurement page.

Prompt classSubqueries shownSources retrievedDomains cited
simple1 (the prompt itself)93
ambiguous1 (the prompt itself)92
longtail3164
multifaceted (3 runs)3, 3, 315, 15, 159, 9, 11
comparison3159
commercial3, plus 3 page fetches in a second step153
rare3157
fresh3234
local3153
product3176
followupnot displayed94
Horizontal bar chart of sources retrieved per prompt class on Perplexity: fresh 23, product 17, longtail 16, multifaceted, comparison, commercial, rare and local 15 each, all with three subqueries; simple, ambiguous and follow-up 9 each with a single search or none shown
Sources retrieved by prompt class, batch 1. Single-search prompts retrieved 9; fan-outs retrieved 15 to 23.

Batch 1 records

Abbreviated view. The CSV and JSON carry every field including timestamps, cited domains, notes and the thread link for each capture.

idrunclasspromptobserved subqueriesnsrcclass
qfo-00011multifacetedWhat is the best CRM for a 50-person law firm concerned about security and migration costbest CRM for law firm 50 employees security · legal CRM migration cost considerations 2025 2026 · top secure CRM platforms for mid-size law firms315OBS
qfo-00012multifaceted(same)best CRM for law firm 50 employees security · legal CRM migration cost considerations · top CRM for mid-size law firms 2025 2026 security compliance315OBS
qfo-00013multifaceted(same)best CRM for law firm 50 employees security · legal CRM migration cost comparison 2025 2026 · law firm CRM security compliance Clio Lawmatics Filevine315OBS
qfo-00021simpleHow tall is the Eiffel TowerHow tall is the Eiffel Tower19OBS
qfo-00031longtailHow do I stop my sourdough starter from developing a grey liquid layer on topsourdough starter grey liquid layer on top what is it · how to prevent hooch sourdough starter grey liquid · sourdough starter gray liquid hooch fix316OBS
qfo-00041comparisonSony WH-1000XM6 vs Bose QuietComfort Ultra for long flightsSony WH-1000XM6 vs Bose QuietComfort Ultra long flights · Sony WH-1000XM6 battery life ANC comfort review 2025 2026 · Bose QuietComfort Ultra battery life ANC comfort long flight review315OBS
qfo-00051commercialBest standing desk under $400 for a small apartmentbest standing desk under $400 2025 2026 · compact standing desk small apartment under $400 · standing desk reviews small space budget315OBS
qfo-00061ambiguousjaguar speedjaguar speed19OBS
qfo-00071rareTraditional dairy dishes served at Tsagaan Sar in Uvs province MongoliaTsagaan Sar traditional dairy dishes Uvs province Mongolia · Mongolian Tsagaan Sar food dairy products Uvs · traditional Mongolian dairy dishes holiday Tsagaan Sar315OBS
qfo-00081freshWhat changed in the most recent iOS updateiOS 26.0 update changes features · Apple iOS latest update September 2026 what's new · iOS 26 release notes changes323OBS
qfo-00091localBest time to visit Great Sand Dunes National Park to avoid crowdsbest time to visit Great Sand Dunes National Park avoid crowds · Great Sand Dunes National Park least crowded months weekdays · Great Sand Dunes National Park crowd patterns seasonal tips315OBS
qfo-00101productIs the DJI Mini 5 Pro worth it over the Mini 4 ProDJI Mini 5 Pro vs Mini 4 Pro differences specs · DJI Mini 5 Pro worth it over Mini 4 Pro review · DJI Mini 5 Pro release date price features317OBS
qfo-00111followupWhich one has better battery life (turn 2 of qfo-0004)not displayednull9OBS*

* Recorded as OBSERVED with fanout_observable false: the observation is that nothing was displayed.

Schema

Columns in the CSV and keys in each JSON record. Lists are pipe-separated in the CSV.

FieldTypeMeaning
idstringStable identifier for the prompt, e.g. qfo-0001. Repeat runs share the id and differ in run.
runintegerRun number for repeated prompts, starting at 1.
promptstringExact prompt text submitted.
prompt_classenumsimple, longtail, multifaceted, comparison, commercial, ambiguous, rare, fresh, local, product, followup.
platformstringperplexity, google-ai-mode, google-ai-overview, chatgpt-search, and so on.
timestamp_utcISO 8601When the test was run, to the minute.
locationstringCountry and region of the tester.
logged_inbooleanWhether the session was signed in.
fanout_observablebooleanWhether the platform displayed any subqueries for this turn.
observed_subquerieslistSubqueries exactly as displayed. Empty when not observable. Direct page fetches are not subqueries and are described in notes instead.
subquery_countinteger or nullLength of the list above; null when not observable.
sources_retrievedinteger or nullThe number the platform reported consulting (Perplexity's "N sources").
sources_citedintegerDistinct domains attached to the final response as citations.
cited_domainslistThose domains, as the platform labelled them.
repeat_runsintegerTotal runs of this prompt in the batch.
repeat_consistencyfloat or nullShare of subqueries identical, word for word, across all runs. 0.33 for qfo-0001 (one of three).
overlap_with_plain_searchfloat or nullShare of cited domains also in the top 10 of a plain web search for the prompt. Not computed in batch 1.
evidence_classenumOBSERVED, RECONSTRUCTED, SYNTHETIC. Batch 1 is entirely OBSERVED.
evidence_filestringThe Perplexity thread the record was captured from. Screenshots will be added under the same field in a later version.
notesstringStep labels, second-step fetches, and anything the fields cannot express.

Versions

Planned for 0.2: ChatGPT search rows for the same eleven prompts; manual AI Mode captures with screenshots (RECONSTRUCTED); plain-search overlap computed for all rows; repeat runs for every prompt.

How to cite

QueryFanout.wiki (2026). Query Fan-Out Observatory: timestamped AI search fan-out tests, version 0.1.0.
https://queryfanout.wiki/data/. Licensed CC BY 4.0.

If you use a single record, cite it by id and run. If you reproduce a subquery list, keep the platform and timestamp with it. The whole point of the dataset is that these are observations tied to a moment, not facts about a system.