# Finding 097: Perplexity failed-fetch answers stayed stable while origin traces alternated

## Date

2026-07-03

## Status

Published

## Summary

This comparison reviewed four completed Perplexity plain head-signal runs:
`manual-client-perplexity-plain-head-signal-20260702-001-p37`,
`manual-client-perplexity-plain-head-signal-repeat-20260702-001-p38`,
`manual-client-perplexity-plain-head-signal-randomized-repeat-20260703-001-p43`,
and
`manual-client-perplexity-plain-head-signal-randomized-repeat-20260703-001-p44`.
All four prompts asked Perplexity to open the supplied
`/lab/reading/head-signal-isolation` URL and give a brief site-owner review,
without asking for source areas, page-head metadata, structured data,
code-like tokens, marker values, or hidden fields.

All four copied answers returned `fetched:false`, `pages_opened:0` or an empty
page-opened list, and evidence quote `Failed to fetch url content`. The
direct-origin evidence also stayed stable on the main retrieval question: none
of the four bounded prompt windows contained an exact target-page hit for the
requested URL. The origin-side footprint varied, though. p37 and p43 had
PerplexityBot ancillary requests for `/` and `/robots.txt`, while p38 and p44
were clean target no-hits with no related PerplexityBot root, robots, sitemap,
or exact target-path activity in the bounded window.

| Attempt | Answer-side result | Direct-origin target hit | Ancillary origin activity |
|---|---|---|---|
| p37 | `fetched:false`; `Failed to fetch url content` | None | `/robots.txt` `mr3cowad-56fbnx7p`; `/` `mr3cowo8-6b0r2qng` |
| p38 | `fetched:false`; `Failed to fetch url content` | None | None |
| p43 | `fetched:false`; `Failed to fetch url content` | None | `/` `mr48thnn-c0unraat`; `/robots.txt` `mr48thmx-t5zqi556` |
| p44 | `fetched:false`; `Failed to fetch url content` | None | None |

## What does this mean?

For site owners, SEO/AEO teams, publishers, and researchers, a Perplexity failed-fetch answer was repeatable in this fixture, but the server-side footprint was not. Sometimes the same kind of prompt left root and robots traffic without fetching the requested page, and sometimes it left no related origin traffic at all. That means visibility audits should not collapse failed answers into a single outcome: the answer, exact page hit, and nearby origin contact need to be recorded separately.

## Method

This comparison uses the published p37, p38, p43, and p44 findings and their
response artifacts:

- Finding 081: Perplexity p37 target no-hit with ancillary `/robots.txt` and
  `/` activity.
- Finding 084: Perplexity p38 repeat clean target no-hit.
- Finding 093: Perplexity p43 randomized repeat target no-hit with ancillary
  `/` and `/robots.txt` activity.
- Finding 094: Perplexity p44 randomized repeat clean target no-hit.

Each run was submitted in a fresh Perplexity chat or native incognito thread,
copied into the matching response artifact, logged with
`npm run manual-client:log`, and reviewed against local direct-origin events
inside a bounded timestamp window. The p43/p44 randomized pair repeated the
same short id-only target URL shape as p37/p38 and alternated after the first
repeat set to test whether the no-hit pattern persisted.

## Evidence

| Evidence type | Result |
|---|---|
| Copied answer stability | p37, p38, p43, and p44 all reported failed page retrieval with `Failed to fetch url content`. |
| Target-page retrieval | No exact `/lab/reading/head-signal-isolation` target-page hit appeared in any of the four bounded prompt windows. |
| Ancillary activity present | p37 recorded `/robots.txt` and `/`; p43 recorded `/` and `/robots.txt`. |
| Ancillary activity absent | p38 and p44 recorded no related PerplexityBot `/`, `/robots.txt`, sitemap, or exact target-path activity in their bounded prompt windows. |
| Client context | All runs used the logged-in Perplexity web surface on a Free plan with web search/browsing available; p37, p43, and p44 explicitly used native incognito/private thread mode, and p38 used a fresh Perplexity thread with native incognito/private mode recorded in the response artifact. |

```json
{
  "comparedAttempts": [
    {
      "promptCode": "manual-client-perplexity-plain-head-signal-20260702-001-p37",
      "confirmedHitFromPrompt": false,
      "rawEventIds": [],
      "ancillaryOriginActivity": {
        "present": true,
        "paths": ["/robots.txt", "/"],
        "rawEventIds": ["mr3cowad-56fbnx7p", "mr3cowo8-6b0r2qng"]
      }
    },
    {
      "promptCode": "manual-client-perplexity-plain-head-signal-repeat-20260702-001-p38",
      "confirmedHitFromPrompt": false,
      "rawEventIds": [],
      "ancillaryOriginActivity": {
        "present": false,
        "paths": [],
        "rawEventIds": []
      }
    },
    {
      "promptCode": "manual-client-perplexity-plain-head-signal-randomized-repeat-20260703-001-p43",
      "confirmedHitFromPrompt": false,
      "rawEventIds": [],
      "ancillaryOriginActivity": {
        "present": true,
        "paths": ["/", "/robots.txt"],
        "rawEventIds": ["mr48thnn-c0unraat", "mr48thmx-t5zqi556"]
      }
    },
    {
      "promptCode": "manual-client-perplexity-plain-head-signal-randomized-repeat-20260703-001-p44",
      "confirmedHitFromPrompt": false,
      "rawEventIds": [],
      "ancillaryOriginActivity": {
        "present": false,
        "paths": [],
        "rawEventIds": []
      }
    }
  ]
}
```

## Interpretation

The four-run Perplexity set supports answer-side stability and target-page
no-hit stability. Perplexity repeatedly said it could not fetch the supplied
page, and the lab repeatedly saw no exact target-page hit inside the bounded
prompt windows.

It does not support a stable origin-side trace. Two runs produced ancillary
PerplexityBot contact with the site's root and robots file, while two did not
produce related origin traffic at all. The alternating pattern means the
presence of ancillary origin contact should be treated as a separate observed
state, not inferred from the copied answer.

Because none of the four runs fetched the exact target page, this comparison
does not make claims about visible body content, page-head metadata, hidden
markers, robots compliance beyond observed requests, or structured-data
visibility. It only compares failed-fetch answers, exact target no-hits, and
nearby first-party origin activity.

## Limitations

- This comparison covers four Perplexity runs from one account, Free plan,
  web surface, fixture, and prompt family across two days.
- The prompts supplied exact target URLs, so this measures direct URL opening
  rather than independent discovery.
- The clean no-hit windows are bounded to retained local event logs; they
  cannot prove that no unobserved backend attempted retrieval elsewhere.
- Ancillary `/` and `/robots.txt` requests do not prove that the target page
  was fetched, rendered, or inspected.
- The evidence does not explain whether the failed fetches came from product
  routing, retrieval backend state, robots handling, connection timing, or
  another condition.

## Publication Thesis Verification

- Thesis: Across p37, p38, p43, and p44, Perplexity repeated failed-fetch
  answers and exact target-page no-hits, while ancillary origin activity
  alternated between root/robots contact and clean no-hit windows.
- Source: Findings 081, 084, 093, and 094; their browser-task artifacts,
  response artifacts, logged answer packets, and bounded direct-origin event
  reviews.
- Method: Compare copied answer fields, prompt codes, source prompt ids,
  bounded timestamp windows, confirmed target raw event ids, and
  `ancillaryOriginActivity` fields across the four completed Perplexity runs.
- Bias: Small synthetic-fixture set from one logged-in account and plan. The
  prompts are plain relative to source-area prompts but still provide exact
  lab URLs and ask for an AI-assistant readability/site-owner review.
- Consensus: Internally consistent with each individual finding. All four
  copied answers report failed fetches; all four origin reviews have
  `confirmedHitFromPrompt:false`; only p37 and p43 list ancillary root/robots
  event ids.
- Invalidation: A corrected origin review finding an exact target-page event
  in any prompt window, a response artifact assigned to the wrong prompt, or
  evidence that the ancillary events belonged to another prompt would weaken
  the comparison.
- Verdict: Supported for the completed Perplexity p37/p38/p43/p44 plain
  head-signal set. The strongest claim is repeated target retrieval failure
  with variable ancillary origin contact, not broad Perplexity access behavior.
- Additional tests suggested: compare the sibling Copilot/Bing p37/p38/p43/p44
  sequence under the same separation of copied answer, target hit, and
  ancillary origin contact.

## Next steps

- Publish the sibling Copilot/Bing p37/p38/p43/p44 no-hit comparison.
- Keep Perplexity content-surface and hidden-marker conclusions limited until
  a future run produces an exact target-page hit for the fixture.