# Finding 083: Plain head-signal prompts split into confirmed hits and target no-hits across clients

## Date

2026-07-02

## Status

Published

## Summary

The plain head-signal comparison reviewed five completed controlled-browser
runs against the `/lab/reading/head-signal-isolation` fixture:

| Client | Attempt | Retrieval result | Origin result | Copied answer scope |
|---|---|---|---|---|
| Claude | `manual-client-claude-plain-page-quality-20260702-001-p36` | `fetched:true`, `pages_opened:1` | Target-page hit `mr31xowy-j9qdrdq1`; ancillary `/robots.txt` event `mr31xohl-eo2hrr9o` | Diagnostic purpose and visible body sentence; no exact marker values |
| ChatGPT | `manual-client-chatgpt-plain-head-signal-20260702-001-p37` | `fetched:true`, `pages_opened:1` | Target-page hit `mr38digu-q1ak63ho` | Diagnostic purpose and visible body text; no exact marker values |
| Gemini | `manual-client-gemini-plain-head-signal-20260702-001-p37` | `fetched:true`, `pages_opened:1` | Target-page hit `mr3ahz4q-7kfpdrht` | Diagnostic purpose and exact visible marker `VISIBLE-QUARTZ-39`; no exact hidden page-head marker values |
| Perplexity | `manual-client-perplexity-plain-head-signal-20260702-001-p37` | `fetched:false`, `pages_opened:0` | No target-page hit; ancillary `/robots.txt` event `mr3cowad-56fbnx7p` and root event `mr3cowo8-6b0r2qng` | Failed target fetch |
| Copilot/Bing | `manual-client-copilot-bing-plain-head-signal-20260702-001-p37` | `fetched:false`, `pages_opened:0` | No target-page hit and no related ancillary origin activity | Failed target fetch |

All five prompts used ordinary site-owner review wording and avoided asking for
source areas, page-head metadata, structured data, code-like tokens, marker
values, or hidden fields. Claude, ChatGPT, and Gemini fetched the exact target
page. Perplexity and Copilot/Bing returned failed-fetch answers without exact
target-page hits; Perplexity still contacted `/robots.txt` and `/`, while
Copilot/Bing showed no related origin activity in the bounded window.

## What does this mean?

For site owners, SEO/AEO teams, publishers, and researchers, the same public page and similar plain review prompt can produce different retrieval outcomes across AI clients. A successful answer, a visible marker quote, a root or robots request, and a failed-fetch answer are different evidence types. The practical takeaway is that AI visibility testing needs both the copied assistant answer and direct server logs before deciding whether a client actually read a page or only contacted the site around it.

## Method

This comparison uses the published p36/p37 findings and their response
artifacts:

- Finding 077: Claude p36 plain page-quality confirmed hit.
- Finding 079: ChatGPT p37 plain head-signal confirmed hit.
- Finding 080: Gemini p37 plain head-signal confirmed hit.
- Finding 081: Perplexity p37 plain head-signal target no-hit with ancillary
  origin activity.
- Finding 082: Copilot/Bing p37 plain head-signal clean target no-hit.

The p37 prompt packet was prepared for ChatGPT, Gemini, Perplexity, and
Copilot/Bing to match the plainer Claude p36 wording style against the same
fixture. Each run was submitted in a fresh client chat or thread, copied into
the matching response artifact, logged with `npm run manual-client:log`, and
reviewed against the local direct-origin event log inside a bounded timestamp
window.

## Evidence

| Evidence type | Result |
|---|---|
| Confirmed target-page hits | Claude `mr31xowy-j9qdrdq1`; ChatGPT `mr38digu-q1ak63ho`; Gemini `mr3ahz4q-7kfpdrht` |
| Target no-hit with ancillary origin activity | Perplexity requested `/robots.txt` (`mr3cowad-56fbnx7p`) and `/` (`mr3cowo8-6b0r2qng`) but did not hit the exact target page |
| Clean target no-hit | Copilot/Bing had no exact target-page hit and no related root, robots, or sitemap activity in the bounded window |
| Exact hidden page-head marker values in copied answers | Not observed for any plain p36/p37 answer |
| Exact visible marker in copied answers | Gemini quoted `VISIBLE-QUARTZ-39`; Claude and ChatGPT described visible body text without copying the exact marker value |

## Interpretation

The comparison supports three separate retrieval states under similar plain
wording:

1. Confirmed target retrieval: Claude, ChatGPT, and Gemini fetched the exact
   target page and produced page-aware answers.
2. Site contact without target retrieval: Perplexity returned a failed-fetch
   answer while PerplexityBot requested `/robots.txt` and `/`.
3. Clean target no-hit: Copilot/Bing returned a failed-fetch answer with no
   related origin activity in the reviewed window.

For the confirmed-hit clients, the copied answers did not expose exact hidden
page-head marker values. That should be scoped as "not observed in the copied
answers," not as proof that the hidden fields were absent from each client's
internal retrieved representation. Gemini additionally copied the exact visible
body marker, showing that visible marker quotation and hidden marker disclosure
also need separate evidence.

The Perplexity result shows why ancillary origin traffic should not be treated
as page-content retrieval. The Copilot/Bing result shows a stricter no-hit
case where both answer text and origin logs align on failed page retrieval.

## Limitations

- Each client result is one run from one account, tier, mode, and timestamp
  window.
- Claude p36 and non-Claude p37 used matched wording style but not identical
  prompt codes or client products.
- The prompts supplied exact target URLs, so the comparison measures direct
  URL opening rather than independent discovery.
- Copied answers describe each client surface's extracted view, not raw HTTP
  bytes or complete internal retrieval state.
- The fixture is synthetic and diagnostic; its visible page text can make the
  body-versus-head setup obvious after retrieval.

## Publication Thesis Verification

- Thesis: Under plain site-owner review wording on the same head-signal
  fixture, Claude, ChatGPT, and Gemini had confirmed target-page hits;
  Perplexity had ancillary origin activity without a target-page hit; and
  Copilot/Bing had a clean target no-hit.
- Source: Findings 077, 079, 080, 081, and 082; their browser-task artifacts,
  response artifacts, logged answer packets, and bounded direct-origin event
  reviews.
- Method: Compare copied answers, `fetched` and `pages_opened` fields,
  prompt-code windows, confirmed target raw event ids, ancillary raw event
  ids, and finding-level limitations across the completed p36/p37 run set.
- Bias: Small synthetic fixture, one account per client surface, one run per
  client condition, and prompt wording that still names AI-assistant
  readability and exact lab URLs.
- Consensus: Internally consistent with the individual findings: confirmed-hit
  clients have matching target events, Perplexity has only ancillary origin
  activity, and Copilot/Bing has no related origin activity.
- Invalidation: A corrected origin review finding missing target hits or
  removing cited target hits, response artifacts assigned to the wrong prompt
  windows, or fixture-serving evidence with different markers during the runs
  would weaken this comparison.
- Verdict: Supported for the completed p36/p37 run set. The strongest claim is
  about observed retrieval states, not general platform behavior or hidden
  field availability.
- Additional tests suggested: prepare a repeat plain head-signal packet with
  randomized client order and at least two same-client repeats for clients that
  produced no-hit outcomes.

## Next steps

- Design a repeat plain head-signal packet for Perplexity and Copilot/Bing to
  test whether their p37 target no-hits are stable under the same plain wording.
- Keep the existing mixed-order ChatGPT URL-shape follow-up separate from this
  cross-client head-signal comparison.