GEO Lab #05: Technical Discoverability
What can a search-engine check actually tell me about whether people will discover my work through AI?
Technical learning record, not an AI-answer experiment. The observations below were collected on November 4–6, 2026. No new ChatGPT or Grok recommendations were measured. This page publishes reviewed aggregates; private raw engine/account evidence is excluded.
What I learned
- A successful query can still leave indexing unknown. One Bing batch returned successful results for all 70 URLs, yet the returned fields did not establish that any page was indexed.
- Access, crawling, indexing and recommendation need separate evidence. Seventy successful requests using a declared OpenAI crawler user agent demonstrate simulated access, not a real OpenAI crawler visit.
- Submission and processing are different outcomes. Google accepted the Sitemap submission; its later processing flag was still pending. Bing separately reported a successful imported feed.
- Keep failures visible. A later joint run hit Bing throttling after 11 successful URL queries. The failed request and 58 skipped positions remain in the history; they were not converted into indexing results.
These are time-bounded observations, not a controlled GEO intervention. The successful ten-second batch and later five-second batch differed in time as well as spacing. They cannot prove that one request interval caused the different outcomes.
Later verified observations
These dated follow-ups extend the record. They do not overwrite the earlier snapshots or establish an AI recommendation effect.
| Time (UTC) | Observation | Verified result | Boundary |
|---|---|---|---|
| 2026-11-06T11:59:17.427043+00:00 | The Google Search Console Sitemaps screen reported "Could not read this sitemap", with a displayed last-read date of November 4 and zero discovered pages. | This is an independently recorded owner-console observation, separate from the API isPending=true response. No corrective resubmission had been made at this observation. | The screen did not expose a specific HTTP or XML error. A successful API submission and a browser HTTP 200 do not establish a successful Google fetch. |
| 2026-11-06T12:57:21.739Z | One corrective Google Sitemap submission was accepted in the owner console after the genuine Lab #05 publication changed the verified Sitemap from 10 to 12 canonical URLs. | The success confirmation followed the submission started at 12:56:01 UTC. This was a separate corrective UI action, not a second initial API PUT or a recurring duplicate submission. | Acceptance does not establish that Google downloaded, processed or indexed the new pages. The earlier could-not-read observation remains in the record. |
| 2026-11-06T13:03:44.549018+00:00 | The expanded Google indexed-version snapshot covered 12 important canonical URLs. | 5 URLs were INDEXED and 7 were NOT_INDEXED. The two new editorial pages are part of this later twelve-URL scope, not the earlier ten-URL baseline. | These are official Google indexed-version observations at this time, not live tests, Bing outcomes or evidence of AI recommendations. |
| 2026-11-06T13:03:44.221335+00:00 | The Google Sitemap API was read after the corrective UI submission. | isPending remained true, lastDownloaded was absent, and the returned error and warning counts were zero. | A successful submission is not confirmed processing. No new download time is inferred from the missing field. |
| 2026-11-06T13:05:02.774161+00:00 | The later combined monitor queried all 72 current canonical content URLs in Bing with ten-second spacing. | 72/72 URL queries succeeded; site access and the Sitemap read also succeeded. The feed remained registered with processing status Success. | All 72 Bing index-membership states remain UNKNOWN. This later success does not erase the earlier five-second throttled run or prove that spacing alone caused recovery. |
| 2026-11-06T13:17:48.178440+00:00 | OpenAI search crawl-eligibility checks expanded to the 72 current canonical content URLs. | 72 configuration-allowed results and 72 successful declared-user-agent simulated fetches were recorded. | Real OAI-SearchBot access and OpenAI index membership remain UNKNOWN; simulated access is not a verified crawler visit. |
| 2026-11-06T13:18:13.900470+00:00 | The full combined monitoring workflow completed successfully and made one real IndexNow notification for 11 genuinely new or updated URLs. | The response was HTTP 202 / KEY_VALIDATION_PENDING. The notification, expanded snapshots and history were saved; no ChatGPT/Grok answer experiment was run. | HTTP 202 means key validation remains pending. Receipt acceptance is not crawling, indexing or AI recommendation evidence; unchanged URLs are not repeatedly submitted merely to obtain another status. |
| 2026-11-06T13:33:18.073Z | Between 2026-11-06T13:19:02.659Z and 2026-11-06T13:33:18.073Z, an owner-authorized assistant used the official Google Search Console interface to request indexing for seven pages. | A fresh indexed-version inspection before each request showed NOT_INDEXED. All seven requests received the interface acknowledgement that the URL had been accepted into the priority crawl queue. | Accepted indexing requests are not evidence that the pages were indexed. The 13:03 UTC snapshot of 5 INDEXED and 7 NOT_INDEXED predates these requests and is not backfilled; this follow-up claims no new post-request indexing result. |
Initial observations, four meanings
Frozen initial scope and time: Google checked 10 important editorial routes; Bing, OpenAI eligibility and the initial IndexNow ledger covered 70 canonical content URLs, including individual-answer archives. These are not interchangeable denominators. The new reader and methods page add two editorial URLs after those snapshots; they were not part of the earlier counts.
| Dimension | Observed result | Interpretation | Boundary |
|---|---|---|---|
| Google indexing | 5 indexed / 5 not indexed, among 10 important URLs | Official indexed-version API results | A Google index result does not establish Bing or ChatGPT indexing. |
| Bing URL information | 70/70 URL queries succeeded in one batch | 70 index-membership states remain UNKNOWN | HTTP 200, IsPage and default dates are not membership verdicts. |
| OpenAI crawl eligibility | 70 configuration-allowed / 70 simulated fetches succeeded | Real crawler visits and OpenAI indexing: UNKNOWN | The requests were made by the monitoring environment, not verified OpenAI source IPs. |
| IndexNow | 0 notification POSTs at the recorded snapshot; 70 unchanged URLs | Key file publicly reachable; remote validation not yet observed | No unchanged URL was re-submitted just to manufacture a receipt. |
Google snapshot: . Bing successful batch started: . The methods appendix keeps the later failed run and separate source times visible.
Ask what each signal proves
- Configuration and access: canonical URLs, indexability directives, effective robots rules and HTTP access establish current technical conditions.
- Discovery and crawling: an engine-associated crawl timestamp or verified request log can show that a system encountered a page. A missing timestamp leaves that evidence unknown.
- Indexing: use the engine's explicit indexed-version verdict when available. Do not infer membership from a successful information query.
- AI retrieval and recommendation: test user questions separately, with repeat runs and original answers. Search-engine indexing alone does not supply this evidence.
The original GitHub Pages URL does not currently expose a confirmed owner-accessible per-request crawler log. Repository Traffic is not a substitute for a Pages access log. Real OAI-SearchBot access therefore remains UNKNOWN.
A small workflow I can keep using
- Save a baseline before making changes. Keep manual observations and official API responses separate; do not backfill later results into an earlier date.
- Reuse one canonical URL inventory. Record which routes belong in the Sitemap and which supplemental archives remain crawlable.
- Keep engines separate. Save Google indexing, Bing information queries and OpenAI eligibility in their own snapshots.
- Stop on throttling or access failures. Retain the first blocking response and every remaining unqueried URL as UNKNOWN with zero attempts.
- Preserve history privately. Raw responses and account evidence stay outside the public Pages repository. Publish only reviewed summaries.
- Notify only after evidence is durable. A completed, successful GitHub run and committed full snapshot precede the first complete-batch notification.
The recurring monitor is configured for Monday 10:17 Asia/Shanghai; GitHub may delay its start. Configuration and a manual workflow dispatch do not prove that the first scheduled occurrence has run. Ordinary Google indexing requests remain manual.
What the failed run changed
| Recorded batch | URL results | Spacing | What remains unknown |
|---|---|---|---|
| Earlier Bing-only batch | 70 succeeded; 0 failed; 0 skipped | 10-second configuration; minimum observed response gap about 10.21 seconds | Indexing and valid page-crawl evidence were not established. |
| Later combined monitor | 11 succeeded; 1 ThrottleHost fault; 58 skipped after the block | 5-second configuration | No results are invented for skipped positions; future quota availability is unknown. |
Bing returned ThrottleHost (official code 5) with HTTP 400. Treating only HTTP 429 as throttling would miss this case. The program now recognizes provider fault codes, stops the batch and records skipped positions. A successful earlier batch does not erase a later failure.
What I would reuse as an independent builder
- Measure one layer at a time. Name the evidence you want before choosing an API or dashboard.
- Use UNKNOWN as a real result. Missing or ambiguous fields should not become a negative indexing verdict.
- Publish useful changes, then notify. A genuine report or corrected learning resource can create an eligible IndexNow change; a test edit, date-only change or repeated old URL does not establish a useful intervention.
- Keep a change timeline. Technical fixes, Sitemap submission, manual indexing requests and later publication are separate events. Their effects cannot be pooled into one causal claim.
The initial snapshot preceded a real IndexNow receipt and a successful combined run; the dated follow-ups above now document those operational checks. Google Sitemap processing, Bing index membership, real OpenAI crawling and remote IndexNow key validation remain unconfirmed. None of these observations establishes recommendation lift.
Methods, sources and publication scopeDownload reviewed public summary
Author: Helen Bird · Observation period: 2026-11-04 to 2026-11-06 (technical observations; no AI-answer experiment) · First published: 2026-11-06 · Last revised: 2026-11-06 · Version: v1.1
Added dated follow-ups: expanded 12/72 URL monitoring, successful combined run, genuine IndexNow receipt with key validation pending, corrective Sitemap submission, and seven acknowledged indexing requests. Initial observations and failures remain unchanged; private raw data remains excluded.
Version metadata · Revision history · Cross-report review decisions · Stage 4 review decisions · Technical maintenance