# Finding 059: ChatGPT retrieved the hidden-signal page again when neutral wording used a short URL

## Date

2026-07-01

## Status

Published

## Summary

The ChatGPT p23 neutral retrieval-stability rerun reused the same
`/lab/reading/hidden-structured-data-conflict` fixture as p21 and p22, but
changed two things from the p22 no-hit: it used a short id-only target URL and
asked for a concise site-owner page-quality note without naming metadata or
JSON-LD. ChatGPT returned `fetched:true`, `pages_opened:1`, quoted
`VISIBLE-PLUM-47`, and said the rendered page view did not directly expose the
meta-description or JSON-LD source fields.

Direct-origin raw event `mr21pcqo-4w1raylc` confirmed the exact target-page
hit from `ChatGPT-User/1.0` inside the bounded
`2026-07-01T12:19:58.749Z` to `2026-07-01T12:22:22.218Z` prompt window. The
operator preflight event for the same attempt occurred before prompt
submission and is not counted as prompt-caused evidence.

## What does this mean?

For site owners and researchers, ChatGPT's earlier neutral no-hit was not necessarily a stable refusal or inability to read this page. With a shorter target URL and similarly neutral site-owner wording, ChatGPT reached the page again and summarized only what was visible in its rendered page view. That keeps the hidden-field conclusion unchanged for this run, but it shows retrieval itself can depend on prompt and URL presentation details.

## Method

- Prepared packet:
  `research/manual-client-runs/manual-client-chatgpt-neutral-stability-20260701-001.prompts.json`.
- Browser task:
  `research/manual-client-runs/browser-tasks/manual-client-chatgpt-neutral-stability-20260701-001-p23.browser-task.json`.
- Response artifact:
  `research/manual-client-runs/browser-tasks/responses/manual-client-chatgpt-neutral-stability-20260701-001-p23.response.json`.
- Logged answer packet:
  `research/manual-client-runs/manual-client-chatgpt-neutral-stability-20260701-001.answers.json`.
- Prompt code:
  `manual-client-chatgpt-neutral-stability-20260701-001-p23`.
- Source prompt:
  `chatgpt-neutral-page-quality-short-url`.

The exact target URL was preflighted before submission and returned `200 OK`
with `VISIBLE-PLUM-47`. The prompt was then submitted in a fresh ChatGPT chat
from the logged-in OpenClaw browser profile. The visible client context showed
the `ChatGPT` selector, account initials `KB`, an `Upgrade` button, and the
`Look something up` control before submission. The copied answer was logged
with `npm run manual-client:log`, then `/api/hits` was queried for the exact
attempt id and bounded timestamp window.

## Observed Result

ChatGPT returned compact JSON with:

- `fetched:true`
- `pages_opened:1`
- visible token `VISIBLE-PLUM-47`
- evidence quote `The page quality code is VISIBLE-PLUM-47.`
- limitation text saying only rendered page content was observed and meta
  description and JSON-LD source were not directly visible

The reviewed direct-origin event was:

| Raw event id | Timestamp | Path | User agent |
|---|---|---|---|
| `mr21pcqo-4w1raylc` | `2026-07-01T12:22:08.099Z` | `/lab/reading/hidden-structured-data-conflict` | `ChatGPT-User/1.0` |

## Interpretation

This result restores target-page retrieval for ChatGPT under neutral wording,
but it does not prove hidden metadata or JSON-LD visibility. ChatGPT's copied
answer explicitly scoped its observation to the rendered page view and reported
only `VISIBLE-PLUM-47`. That matches ChatGPT p21's visible-only result while
contrasting with the p22 clean no-hit.

The difference from p22 is not isolated to one variable. p23 changed both
wording and URL shape: p23 used a short id-only target URL, while p22 used the
current default non-Claude correlation URL with multiple query parameters.
Prepared p24 and p25 tasks should separate these factors by testing full
correlation metadata and a second short-URL neutral framing.

## Limitations

- This is one ChatGPT run in one logged-in account context and one fresh chat.
- The test supplied the exact target URL, so it measures direct URL opening,
  not independent discovery.
- The p23 result differs from p22 in both wording and URL shape, so it should
  be treated as retrieval-stability evidence rather than a final explanation
  of the p22 no-hit.
- The answer reports rendered-view limitations. It does not prove that other
  ChatGPT modes, models, account tiers, or future retrieval paths cannot see
  metadata or JSON-LD.

## Publication Thesis Verification

- Thesis: ChatGPT p23 restored retrieval of the hidden-signal target under
  neutral wording with a short id-only URL and reported only the visible page
  code, not meta-description or JSON-LD codes.
- Source: The p23 browser-task artifact, response artifact, logged answer
  packet, `/api/hits` raw event `mr21pcqo-4w1raylc`, and the preflight record
  for the exact target URL.
- Method: Compare copied model JSON with the bounded direct-origin event
  window, excluding the preflight event because it occurred before prompt
  submission.
- Bias: Single-run product behavior can change with model routing, account
  state, client-side search/tool availability, prompt wording, and target URL
  shape.
- Consensus: Consistent with ChatGPT p21 visible-only retrieval and contrary
  to p22's no-hit outcome, which is why p24 and p25 should be run before
  explaining the transition.
- Invalidation: A corrected event review that removes `mr21pcqo-4w1raylc`
  from the prompt window, a response metadata mismatch, fixture-serving error,
  or reruns showing a different stable pattern would weaken the finding.
- Verdict: Supported for this p23 run only. It should be used to motivate the
  remaining neutral stability reruns, not as a broad platform claim.
- Additional tests suggested: run p24 to test neutral wording with full
  correlation query metadata, then run p25 to test a second neutral short-URL
  wording.

## Next steps

- Run p24 from the same packet in a fresh ChatGPT chat.
- Run p25 from the same packet in a fresh ChatGPT chat.
- Compare p22-p25 after all three stability reruns have logged answers and
  origin review.