# Finding 091: ChatGPT fetched both full-correlation and short-id URLs in the mixed packet ## Date 2026-07-03 ## Status Published ## Summary The completed `manual-client-chatgpt-neutral-url-shape-mixed-20260702-001` packet alternated URL shapes under matched neutral readability wording: | Attempt | URL shape | Result | Pages opened | Raw event id | |---|---|---:|---:|---| | p39 | full-correlation query string | `fetched:true` | 2 | `mr3tu40o-5j6n23nt` | | p40 | short id-only query string | `fetched:true` | 1 | `mr3vxsp5-0t1do4up` | | p41 | full-correlation query string | `fetched:true` | 1 | `mr3y3rob-z4q8fk3y` | | p42 | short id-only query string | `fetched:true` | 1 | `mr40985g-99xvskmd` | All four runs were submitted in fresh ChatGPT Temporary Chats from the logged-in OpenClaw-controlled browser profile. All four copied answers reported the visible `VISIBLE-PLUM-47` marker or visible page text, and none reported exact hidden meta-description or JSON-LD marker values. Direct-origin review found one `ChatGPT-User/1.0` target-page hit in each bounded prompt window and no subresource, tracking-pixel, JavaScript beacon, root, robots, or sitemap activity attached to those prompt-caused page views. ## What does this mean? For site owners, SEO/AEO teams, publishers, and researchers, this packet shows that ChatGPT could open the same page whether the supplied URL carried a full tracking-style query string or only a short id parameter. In these four runs, simplifying the URL did not stop retrieval and did not make hidden metadata or structured-data marker values appear in the copied answer. The result is useful evidence about direct URL opening under this prompt family, but it should not be stretched into a universal rule about every ChatGPT mode or every URL shape. ## Method - Prepared packet: `research/manual-client-runs/manual-client-chatgpt-neutral-url-shape-mixed-20260702-001.prompts.json`. - Logged answer packet: `research/manual-client-runs/manual-client-chatgpt-neutral-url-shape-mixed-20260702-001.answers.json`. - Response artifacts: `research/manual-client-runs/browser-tasks/responses/manual-client-chatgpt-neutral-url-shape-mixed-20260702-001-p39.response.json`, p40, p41, and p42. - Run-level findings reviewed: Findings 087, 088, 089, and 090. The comparison reviewed the four completed response artifacts, the logged answer packet, and the numbered run-level findings. The packet was designed to interleave full-correlation and short id-only URLs: p39 and p41 carried the complete manual-client query parameters, while p40 and p42 carried only the attempt id in the target URL. The prompt text still included the attempt id, source prompt id, prompt code, AI system, model id, task, and target URL for all four runs. ## Observed Result The response artifacts record these bounded windows: | Attempt | Window | Confirmed target hit | |---|---|---| | p39 | `2026-07-02T18:16:24Z` to `2026-07-02T18:18:10Z` | `mr3tu40o-5j6n23nt` | | p40 | `2026-07-02T19:16:14Z` to `2026-07-02T19:16:35Z` | `mr3vxsp5-0t1do4up` | | p41 | `2026-07-02T20:15:37Z` to `2026-07-02T20:17:26Z` | `mr3y3rob-z4q8fk3y` | | p42 | `2026-07-02T21:16:18Z` to `2026-07-02T21:17:28Z` | `mr40985g-99xvskmd` | All four response artifacts have `confirmedHitFromPrompt:true`, `confirmationStatus:"confirmed_origin_hit"`, and empty ancillary-origin activity. The visible client context stayed consistent across the packet: ChatGPT Free account tier, ChatGPT model selector text, native Temporary Chat mode, and no explicit browsing/search label shown. The copied answers stayed in the same content band. The full-correlation p39 and p41 answers explicitly said meta-description or JSON-LD code was not directly observable in the accessible page view. The short id-only p40 and p42 answers quoted visible page text and said they did not infer or report hidden metadata or JSON-LD beyond the directly accessible view. ## Interpretation Inside this packet, URL shape did not separate ChatGPT retrieval outcomes. Two full-correlation URLs and two short id-only URLs all produced prompt-window target-page hits from `ChatGPT-User/1.0`. The one visible difference was answer-side page count: p39 reported `pages_opened:2`, while p40-p42 reported `pages_opened:1`; direct-origin review still found one exact lab target-page event for each attempt. The comparison strengthens the narrower claim that, for this account, mode, fixture, wording, and time window, the short id-only URL was sufficient for ChatGPT to open the supplied target. It does not prove that URL parameters never matter, that ChatGPT used the same internal retrieval path for both URL shapes, or that hidden page-head fields were absent from the fetched representation. This result should be compared with the earlier separate p29-p31 full-correlation trio and p33-p35 short id-only trio before making a broader URL-shape statement across all completed ChatGPT neutral readability repeats. ## Limitations - All four runs used one ChatGPT account context, the Free tier, and native Temporary Chat mode on one day. - The test supplied the exact target URL, so this measures direct URL opening, not independent discovery. - The short id-only runs still included source prompt and prompt code metadata in the submitted prompt text, just not in the target URL query string. - The copied answers are evidence about surfaced answer content, not complete evidence of every field available inside ChatGPT's internal retrieved view. - The packet contains two runs per URL shape, so it is stronger than a single run but still limited. ## Publication Thesis Verification - Thesis: In the completed p39-p42 mixed-order packet, ChatGPT fetched both full-correlation and short id-only target URLs under matched neutral readability wording, and the copied answers stayed visible-page scoped. - Source: Findings 087-090; the p39-p42 response artifacts; the logged answer packet; local raw events `mr3tu40o-5j6n23nt`, `mr3vxsp5-0t1do4up`, `mr3y3rob-z4q8fk3y`, and `mr40985g-99xvskmd`. - Method: Compare URL shape, copied answer fields, `confirmedHitFromPrompt`, raw event ids, bounded timestamp windows, and ancillary-origin activity across the four response artifacts. - Bias: Product routing, account state, Temporary Chat implementation, prompt wording, cached state, and source metadata inside the prompt can all affect retrieval and answer content. - Consensus: Consistent with the run-level findings and with earlier ChatGPT neutral readability repeats that also produced `ChatGPT-User/1.0` target hits and visible-marker copied answers. - Invalidation: A corrected event review that removes any of the four raw events from their prompt windows, a response metadata mismatch, evidence of prompt contamination, or a fixture-serving error would weaken this comparison. - Verdict: Supported for the p39-p42 mixed-order packet. Broader URL-shape claims should wait for a comparison against the p29-p31 and p33-p35 repeat trios. - Additional tests suggested: compare p29-p31, p33-p35, and p39-p42 together; later repeat the same mixed-order design on another day or another ChatGPT mode if supported. ## Next steps - Compare the completed p29-p31 full-correlation trio, p33-p35 short id-only trio, and p39-p42 mixed-order packet. - Keep future URL-shape findings explicit about direct URL opening versus discovery or search-index behavior.