Audience Intel
Read my page
Evidence pack

Evidence, captured.

A screenshot of each Google result and each assistant's answer as a card naming the model, the time and whether web search was on, with whether you and your rivals appear. Real respondents can add their own captures to the same pack.

Start with a benchmark

Built inside Insight Loop · captures spend credits

What a pack holds

One question, five captures.

For each buyer question: the Google results page and four assistants' answers, each a card that says when, which model, with or without web search, and who it names.

Illustration

Q · [a buyer’s question, as typed]

GoogleDD MMM YYYY · HH:MM UTC

Licensed search API · the vendor’s own screenshot

Sponsored

[rival]

AI Overview[source][source]
  1. 1
  2. 2[rival]
  3. 3
  4. 4[your business]

Positions, ads, the AI Overview’s sources and map results are read off the capture.

ChatGPT via APIDD MMM YYYY · HH:MM UTC
model openai/gpt-5.4-miniweb search on

[rival]

[your business]

you appearrival appearssha-256 · ····························
Claude via APIDD MMM YYYY · HH:MM UTC
model anthropic/claude-sonnet-5.5web search on

[rival]

you: not namedrival appearssha-256 · ····························
Gemini via APIDD MMM YYYY · HH:MM UTC
model google/gemini-3.5-flashweb search on

[your business]

you appearrivals: not namedsha-256 · ····························
Grok via APIDD MMM YYYY · HH:MM UTC
model x-ai/grok-4.3web search off · reason given
you: not namedrivals: not namedsha-256 · ····························
Illustration of one question’s captures. Not a real capture: the business, the rivals, the sources and the answers are placeholders.
  1. The timeEvery capture is stamped in UTC.
  2. The modelThe model asked for and the model the API said answered, marked “via API”. Not claimed to be the free app.
  3. Web search on or offWhether the provider's own web search ran for this answer. When it did not, the card says why.
  4. You and your rivalsWhere each is named, highlighted in the answer: green for you, amber for a rival. Or not named at all.
  5. A fingerprintThe answer is kept verbatim, with its SHA-256 on the card, so the card can be checked against the record.
Real respondents

Real people, their own screens.

The same questions, asked by people in their own apps. Their screenshots go into the same pack, labelled as theirs.

The study
Run on Prolific. Quoted per study before it launches; nothing launches without approval.
The capture
Each respondent searches or asks in their own Google or assistant app, takes a screenshot and adds it to the same pack.
Their privacy
They crop out account names and avatars and tick that they did. Images are re-encoded on upload, so location and camera data are dropped.
One each
One submission per person. The participant id is stored only as a keyed hash.
Side by side
Each question shows the API captures and the people's captures together, with how many answered. With too few answers, it says so.
Cost
Quoted, per study, before anyone is asked.
Illustration
account name cropped

[rival]

Captured by a real person · [app] · [date]

Illustration of a respondent’s screenshot as the pack labels it. Not a real capture.
How to get one

Benchmark first, then capture.

A pack is built inside Insight Loop from a finished comparison, so it knows who your rivals are. Nothing is captured from this site.

  1. 01

    Run a benchmark

    Your page and up to three rivals, read by the same 150 agents. Free account, no card.

  2. 02

    Open it in Insight Loop

    The finished run becomes a workspace, with the evidence sheet one click away.

  3. 03

    Capture the evidence

    Buyer questions drafted from your page, editable before anything runs. Captures spend credits, under a cost cap per pack.

  4. 04

    Take it with you

    Download the pack as one file: every card, the data behind it and an index that opens offline.

A pack starts from a benchmark

Run your page and your rivals first. When it is done, the run opens in Insight Loop and the pack is built there.

Run a benchmark

A capture is a timestamped record of what was shown for one question at one time. It is not a ranking, and it says nothing about what buyers do next.

Start with a benchmark.

Your page and up to three rivals, read by the same 150 agents. The evidence pack is built from it.

Run a benchmarkSee pricing