This is what a Full Test tells you
A real test, run on 28 September 2026: AI shopping agents on ChatGPT, Gemini, Perplexity worked through the same customer request 15 times. The business's name and website are hidden; everything else, including the competitors AI chose instead, is exactly as recorded.
AI picked A London seafood restaurant outright in 1 of 15 shopping runs.
You were found in 14 of 15 runs but lost most of them before the pick. J. Sheekey won 4 runs, more than you.
Agent access check
What a legitimate AI shopping agent could and couldn't read during this test. Blocks are recorded, never worked around.
| Your page | Visits | Status |
|---|---|---|
| A London seafood restaurant | 1 | Readable |
| A London seafood restaurant | 1 | Readable |
Other sites that turned the agents away (they then relied on other sources): thefork.co.uk (4), opentable.co.uk (4), opentable.com (2).
Where you dropped out
Every run passes through the same stages. This is how many of the 15 reached each one with you still in it.
How all 15 runs ended
Not every run ends in one winner. We count every outcome.
| Picked you | 1 |
| Picked a competitor | 14 |
| Picked nobody | 0 |
| Stopped without deciding | 0 |
| Didn't finish | 0 |
Who AI chose instead
The 14 runs a competitor won outright.
Named as the place or brand the agent chose; a park's operator or a listing site is folded into the place itself.
Each AI behaves differently
| Assistant | Found | Picked | Winner most often |
|---|---|---|---|
| ChatGPT | 5 / 5 | 0 | Behind |
| Gemini | 5 / 5 | 1 | J. Sheekey |
| Perplexity | 4 / 5 | 0 | Applebee's |
Where AI thinks you fit
| Customer brief | Found | Picked |
|---|---|---|
| Brief 1 | 14 / 15 | 1 |
Brief 1: “Looking for a good fish restaurant in London for 6 people on a Saturday might with a budget of up to GBP 150 per head”
What the agents looked for and couldn't find
Taken from what the agents actually did: the pages they opened, what they searched for, and what they couldn't confirm. Observed means it happened in the recorded runs. Our read is our interpretation.
These are the decision points we saw in this test. Fixing them makes a win more likely but doesn't guarantee one. Rerun the same test afterwards to see what changed.
What the agents searched for instead
| Agent searched | Times |
|---|---|
| “London fish seafood restaurants Saturday dinner six people menu prices” | 1 |
| “site:thefork.co.uk London seafood restaurant Saturday dinner menu £ per person six” | 1 |