# Finding 065: Claude's source-area run exposed multiple head signals but not JSON-LD

## Date

2026-07-01

## Status

Published

## Summary

Claude was prompted in a fresh native incognito chat with
`manual-client-claude-metadata-isolation-20260701-001-p28`, the source-area
boundary variant of the Claude metadata-isolation packet. The target fixture
keeps visible body text separate from page-head signals: visible body marker
`VISIBLE-QUARTZ-39`, meta-description marker `META-CEDAR-71`, OpenGraph title
marker `OG-COBALT-23`, OpenGraph description marker `OG-RUBY-62`, canonical
marker `CANONICAL-PEAR-54`, and JSON-LD marker `SCHEMA-MINT-88`.

Claude returned `fetched:true`, `pages_opened:1`, quoted the visible body
marker, meta-description marker, OpenGraph title and description markers, and
canonical marker, and said its fetched view exposed the page title, robots, and
viewport metadata as well. Claude also said no JSON-LD structured data block
was present in its fetched content. The visible Claude UI fetch label showed
`Fetched: Head signal OpenGraph title code: OG-COBALT-23`, and direct-origin
raw event `mr2c93wu-i61fpyko` confirmed a Claude-User target-page hit inside
the bounded prompt window.

## What does this mean?

For site owners, SEO/AEO teams, and researchers, this run shows that Claude's page-opening view can expose several non-visible page-head signals in addition to normal body text, including description, OpenGraph, and canonical fields. At the same time, Claude's extracted view did not surface the page's JSON-LD marker even though the served page contained it, so different head-adjacent sources should not be treated as equally visible to every assistant retrieval tool.

## Method

- Controlled-browser task:
  `research/manual-client-runs/browser-tasks/manual-client-claude-metadata-isolation-20260701-001-p28.browser-task.json`.
- Prompt packet:
  `research/manual-client-runs/manual-client-claude-metadata-isolation-20260701-001.prompts.json`.
- Response artifact:
  `research/manual-client-runs/browser-tasks/responses/manual-client-claude-metadata-isolation-20260701-001-p28.response.json`.
- Logged answer packet:
  `research/manual-client-runs/manual-client-claude-metadata-isolation-20260701-001.answers.json`.
- Prompt window:
  `2026-07-01T17:15:45Z` to `2026-07-01T17:17:56Z`.

The operator first preflighted the exact public target URL and confirmed a 200
response containing the expected visible body marker, meta-description marker,
OpenGraph markers, canonical marker, and JSON-LD marker. The Claude run then
used a fresh native incognito chat under the Free plan with the visible model
selector `Sonnet 5 Medium`.

## Evidence

| Evidence | Result |
|---|---|
| Model answer | `fetched:true`, `pages_opened:1`; visible marker `VISIBLE-QUARTZ-39`; meta-description marker `META-CEDAR-71`; OpenGraph title marker `OG-COBALT-23`; OpenGraph description marker `OG-RUBY-62`; canonical marker `CANONICAL-PEAR-54`; no JSON-LD observed in Claude's fetched content. |
| Claude UI fetch label | Showed `Fetched: Head signal OpenGraph title code: OG-COBALT-23`. |
| Direct-origin target-page event | `mr2c93wu-i61fpyko` at `2026-07-01T17:17:25.921Z` from `Claude-User/1.0` on `/lab/reading/head-signal-isolation`. |

## Interpretation

The p28 source-area boundary prompt produced the strongest marker-level
metadata result in the p26-p28 Claude metadata-isolation sequence. Unlike p26
and p27, the copied p28 answer explicitly quoted the meta-description,
OpenGraph description, and canonical markers in addition to the visible body
marker and UI-visible OpenGraph title label.

The JSON-LD result remains separate. The served fixture included the JSON-LD
marker `SCHEMA-MINT-88` during preflight, but Claude said no JSON-LD structured
data block was present in the fetched content. This should be read as a limit
of Claude's extracted view in this run, not as proof that the live page lacked
JSON-LD.

## Limitations

- This is one Claude run from one account, tier, model selector, and time
  window.
- The prompt supplied the exact target URL, so this measures direct URL
  opening, not independent discovery.
- The answer describes Claude's `web_fetch` extracted view, not raw HTTP bytes
  or a browser-rendered DOM.
- The prompt explicitly asked for page areas and code-like tokens, so it is
  not a neutral production-user wording test.
- A non-blocking `Claude Fable 5 is currently unavailable` notice appeared in
  the UI while the selected `Sonnet 5 Medium` response completed.

## Publication Thesis Verification

- Thesis: In the p28 source-area head-signal run, Claude fetched the target
  page and surfaced visible body text plus multiple page-head marker values,
  while its extracted view did not surface JSON-LD.
- Source: The p28 browser-task artifact, response artifact, logged answer
  packet, browser transcript, exact-target preflight output, and direct-origin
  raw event `mr2c93wu-i61fpyko`.
- Method: Compare the copied Claude answer and visible browser transcript with
  exact attempt id, prompt code, timestamp window, and raw server-side events.
- Bias: Single-account, single-run evidence. The prompt asks directly for page
  areas and code-like tokens, which can encourage marker reporting when the
  extraction view exposes them.
- Consensus: Consistent with Findings 063 and 064 on confirmed Claude
  retrieval and head-metadata context exposure, while adding stronger
  marker-level evidence for meta-description, OpenGraph description, and
  canonical fields.
- Invalidation: A response-artifact mismatch, corrected origin review removing
  the Claude-User target-page hit, or fixture-serving evidence showing the
  wrong head tags during the run would weaken this finding.
- Verdict: Supported for the p28 run. It confirms retrieval and reported
  source-area/head-signal exposure, while keeping JSON-LD scoped as not
  observed in Claude's fetched view.
- Additional tests suggested: publish a p26-p28 comparison and design a neutral
  follow-up that tests whether the same markers surface without asking for
  page areas or code-like tokens.

## Next steps

- Publish a scoped p26-p28 Claude metadata-isolation comparison.
- Prepare a neutral Claude follow-up that asks for ordinary site-quality
  feedback without naming page-head areas or code-like tokens.