# Personalization context you cannot see

> Part of what an agent weighs belongs to the shopper, not to your catalog: remembered preferences, custom instructions and prior activity, which means two people asking the same question can get different answers for reasons no merchant controls.

<!-- Canonical: https://agentmint.net/signals/personalization-context/ · Signal in the AgentMint.net Signals Database · Evidence: spec-fact (Very high confidence) · Last verified 2026-09-18 · Evidence types are load-bearing: spec-fact | reported | hypothesis (no measured). -->

- **Affects (funnel stage):** Choice
- **Evidence:** spec-fact
- **Confidence:** Very high
- **Impact:** 2/3
- **Status:** active
- **Last verified:** 2026-09-18

## Per-platform status

Undocumented platforms show "unknown" - honest unknowns, never invented.

| Platform | Status | Notes | Source |
| --- | --- | --- | --- |
| ChatGPT | supported | OpenAI documents that ChatGPT considers the query and context such as Memory or Custom instructions, and expects to personalize merchant rankings further over time. | [link](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search) (verified 2026-08-28) |
| Gemini / Google AI Mode | unknown | Not documented by this platform. | - |
| Perplexity | unknown | Not documented by this platform. | - |
| Copilot | unknown | Not documented by this platform. | - |
| Claude | unknown | Not documented by this platform. | - |

## Evidence and sources

- [Shopping with ChatGPT Search (OpenAI Help Center)](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search) (updated approximately 2026-08-28)

## How to act on this

You cannot set this, so treat it as measurement hygiene instead. Test from clean sessions with memory off where the engine allows it, run enough repetitions that personal context averages out, and stop reading a single session as evidence that anything changed.

### Handbook chapters
- [How to Test Agent Selection](https://agentmint.net/handbook/how-to-test-agent-selection/)

### Related signals
- [Defined agent win-rate measurement](https://agentmint.net/signals/agent-win-rate-measurement/)
- [Re-test after every major model update](https://agentmint.net/signals/retest-on-model-updates/)
- [Audit position and ordering effects per engine](https://agentmint.net/signals/position-bias-audit/)

## Notes

This signal is here because it explains variance a merchant would otherwise misread. Someone who sees their product recommended on Monday and absent on Tuesday will look for a ranking change, and the cause may be that the two sessions carried different shopper context. Amazon documents a similar dependence on individual shopping activity for its own assistant. Neither is a lever; both are reasons to distrust a sample of one.