# Finding 070: ChatGPT repeated the full-URL neutral readability hit three times ## Date 2026-07-01 ## Status Published ## Summary Findings 067, 068, and 069 compare three same-condition ChatGPT neutral readability repeats against `/lab/reading/hidden-structured-data-conflict`. All three runs used the p24-style full correlation URL shape, the same neutral readability task family, a fresh ChatGPT Temporary Chat, and the logged-in OpenClaw browser profile. Each run returned `fetched:true`, `pages_opened:1`, reported visible token `VISIBLE-PLUM-47`, and had a direct-origin `ChatGPT-User/1.0` target-page hit inside the bounded prompt window. The repeated content result stayed narrow. ChatGPT did not report the fixture's meta-description token `META-AMBER-16` or JSON-LD token `SCHEMA-INDIGO-82` in any of the three copied answers, even though operator preflight confirmed those served fixture markers before prompt submission. ## What does this mean? For site owners and researchers, this repeat series makes ChatGPT's full-URL page-opening behavior look less fragile than a single run suggested. In this measured setup, three fresh Temporary Chats opened the same long tracking-style URL and answered from visible page text, while still not exposing the hidden metadata or structured-data markers in the copied answer. The practical lesson is to rerun retrieval tests before treating one no-hit as stable behavior, and to keep retrieval success separate from which page fields the answer actually uses. ## Method This comparison reviews the published same-condition repeat findings: - p29: Finding 067, `manual-client-chatgpt-neutral-repeat-20260701-001-p29`. - p30: Finding 068, `manual-client-chatgpt-neutral-repeat-20260701-001-p30`. - p31: Finding 069, `manual-client-chatgpt-neutral-repeat-20260701-001-p31`. - Response artifacts: `research/manual-client-runs/browser-tasks/responses/manual-client-chatgpt-neutral-repeat-20260701-001-p29.response.json` through `research/manual-client-runs/browser-tasks/responses/manual-client-chatgpt-neutral-repeat-20260701-001-p31.response.json`. - Logged answer packet: `research/manual-client-runs/manual-client-chatgpt-neutral-repeat-20260701-001.answers.json`. - Prompt packet: `research/manual-client-runs/manual-client-chatgpt-neutral-repeat-20260701-001.prompts.json`. Each run preflighted the exact target URL, submitted the prompt in a fresh ChatGPT Temporary Chat, logged the copied answer with `npm run manual-client:log`, and reviewed direct-origin evidence by prompt code, attempt id, raw event ids, and bounded timestamp window. Operator preflight requests are treated as fixture-readiness checks, not prompt-caused AI-client behavior. ## Comparison Matrix | Prompt | ChatGPT answer | Direct-origin evidence | Field implication | |---|---|---|---| | p29 | `fetched:true`, `pages_opened:1`; reported `VISIBLE-PLUM-47`. | Raw event `mr2gm1ag-c3itq40v` from `ChatGPT-User/1.0`. | Visible-token answer; meta-description and JSON-LD tokens not reported. | | p30 | `fetched:true`, `pages_opened:1`; reported visible page text including `VISIBLE-PLUM-47`. | Raw event `mr2iptim-5hq0zuxp` from `ChatGPT-User/1.0`. | Visible-token answer; meta-description and JSON-LD tokens not reported. | | p31 | `fetched:true`, `pages_opened:1`; reported visible readability and purpose text. | Raw event `mr2kwg6x-dpw37d24` from `ChatGPT-User/1.0`. | Visible-token answer; meta-description and JSON-LD tokens not reported. | ## Interpretation The p29-p31 set completes the same-condition follow-up proposed by Finding 062. It supports the narrower conclusion that the earlier p22 clean no-hit was not a stable inability to open this fixture or a consequence of full correlation URL shape alone. When p24, p29, p30, and p31 are considered together, four full-correlation URL attempts produced confirmed ChatGPT target hits under neutral site-owner or readability wording. The answer-field conclusion is separate from the retrieval conclusion. In all three repeat runs, ChatGPT copied or summarized visible page text and did not report `META-AMBER-16` or `SCHEMA-INDIGO-82`. That supports a visible-answer pattern for these runs, not a universal claim about all ChatGPT retrieval components, models, modes, or future product states. The series also shows why the lab tracks raw origin evidence and copied answer content separately. A confirmed target-page hit proves the AI-client path contacted the page in the bounded prompt window; it does not by itself prove which HTML fields were exposed to or used by the answer. ## Limitations - This comparison covers three ChatGPT repeats from one logged-in account context, one native Temporary Chat mode, and one fixture. - The prompt supplied the exact target URL, so this measures direct URL opening rather than independent discovery. - The repeats hold URL shape and task family steady, but ordinary product routing and retrieval backends may still vary across time. - The copied answers report what the ChatGPT client exposed in its response; they do not prove what every underlying retrieval component could inspect. - Origin review used bounded prompt windows. Delayed fetches outside those windows would require separate review before changing these conclusions. ## Publication Thesis Verification - Thesis: ChatGPT repeated the p24-style full-URL neutral readability target hit three times in p29-p31, and all three copied answers stayed visible-only for marker evidence. - Source: Findings 067, 068, and 069; their response files; the logged answer packet; browser-task artifacts; and direct-origin raw events `mr2gm1ag-c3itq40v`, `mr2iptim-5hq0zuxp`, and `mr2kwg6x-dpw37d24`. - Method: Compare copied model answers with exact prompt codes, confirmation statuses, raw event ids, URL shapes, task framing, and bounded timestamp windows, while excluding operator preflight requests from prompt-caused evidence. - Bias: The result is single-client, single-account, and short-window. ChatGPT retrieval behavior may vary by tier, model selector, region, browsing tool state, prompt wording, target URL, and time. - Consensus: Consistent with Finding 062's p24 full-correlation confirmed hit and p23-p25 visible-only answers, and with the individual thesis-verification sections for Findings 067, 068, and 069. - Invalidation: A corrected event review that removes any p29-p31 raw event from its prompt window, a response artifact mismatch, a fixture-serving error, or later same-condition repeats showing stable no-hit behavior would weaken the comparison. - Verdict: Supported for this three-run repeat series. The comparison supports repeated full-URL retrieval plus visible-only copied answers when retrieval succeeded; it does not identify one definitive cause for the earlier p22 no-hit. - Additional tests suggested: vary one factor at a time, such as URL length, neutral wording, fixture surface, or native private-mode state, before making broader ChatGPT retrieval claims. ## Next steps - Design a matched short-URL repeat packet to compare against the p29-p31 full-correlation URL repeat series. - Keep future summaries split into retrieval status, origin evidence, and answer-visible field exposure.