Start with a question, then explore the research and its sources.
The experiments, technical observations and research review document what each source measures and where its conclusions apply.
This new index is a content-organization practice suggested by the creator audit. Whether it improves AI discovery or recommendations remains untested.
Hands-on guide
Does AI search recommend my product?
Choose one buyer question, record a baseline, and repeat after a documented change.
GEO Lab #01 · AI Recommendations
How do AI search engines choose sources?
Repeated GEO questions to ChatGPT Search and Grok show differences in recommended entities, cited sources and repeat overlap. Citation, mention and recommendation are separate observations.
12 selected answers; three questions, two consumer interfaces, two repeats. Model and timing changes limit repeat comparisons.
Author: Helen Bird · Experiment: 2026-10-29 to 2026-10-30
First public report: 2026-10-30 · Last revised: 2026-11-06 · Version: 1.4
Original GEO Lab #01 post
GEO Lab #02 · Content Discovery
Why can’t AI find my public content?
Direct URL access, exact-quotation search and unbranded natural discovery test different retrieval tasks. A successful direct read does not imply successful natural discovery.
Six selected A/B/C answers; all ten submissions and technical/capture failures preserved. A qualified manual replacement and later native-search observations remain separate.
Author: Helen Bird · Experiment: 2026-10-31
First public report: 2026-11-01 · Last revised: 2026-11-06 · Version: 1.4
GEO Lab #03 · Prompt Sensitivity
How reliable are GEO measurements?
Repeated and reworded questions explore how measurement advice, cited domains and recommended tools change. Matched comparisons and normalization sensitivity clarify the limits of one-shot scores.
24 selected answers; two intents, two wordings, two environments, three repeats. OpenAI Responses API + web search is not consumer ChatGPT Search.
Author: Helen Bird · Experiment: 2026-11-02
First public report: 2026-11-02 · Last revised: 2026-11-06 · Version: 1.5
GEO Lab #04 · Creator Benchmark
Which independent GEO creators get recommended?
New answers test three historically selected creator families. Actual cited-page audits separate content organization, author attribution, evidence quality and untested practices.
18 selected answers, three candidate creators and eight cited-page audits. Question-level n=3 results are primary; personal-name visibility is separate from associated content assets.
Author: Helen Bird · Experiment: 2026-11-03
First public report: 2026-11-03 · Last revised: 2026-11-06 · Version: 1.2
GEO Lab #05 · Technical Discoverability
What can technical checks prove about discoverability?
A technical learning record separates Google indexing, Bing information queries, OpenAI crawl eligibility and IndexNow notifications. Successful access or API queries do not establish AI recommendations.
Reviewed aggregates only: no new AI-answer experiment. Private raw engine/account evidence is excluded; the earlier snapshots cover 10 or 70 URLs, before these new pages were published.
Author: Helen Bird · Observations: 2026-11-04 to 2026-11-06 (technical observations; no AI-answer experiment)
First public report: 2026-11-06 · Last revised: 2026-11-06 · Version: 1.1
GEO Lab #06 · Content Optimization for AI Search
Which content changes are worth making for AI search?
An evidence review for indie developers and solo founders compares seven GEO studies on relevance, product facts, rewriting, and formatting.
The review distinguishes citation in controlled tests, real-world associations, and the limited evidence for lasting gains from live webpage edits.
Research review: 2026-11-12
GEO Lab #07 · Brand Authority
What makes a creator worth recommending?
Compare which creators get recommended for tutorials or reproducible experiments, then use the evidence to choose improvements a solo builder can make.
60 existing ChatGPT and Grok web answers; three questions and ten answers per question and interface. The release was verified, but its effect on AI recommendations has not been measured.
Research review: 2026-11-13
GEO Lab Interim Review · Part 1
Which findings held up after more runs?
A checkpoint in the ongoing GEO Lab. Part 1 revisits ten original prompts with 300 new answers and shows what changes when three or ten runs are kept.
GEO Lab Interim Review · Part 2
How does question framing change recommendations?
Six questions and 120 web-search answers track how resource and creator prompts change recommendations, source lists and author naming.
Use the evidence
- Read by user intent. Results from different tasks should not be pooled into a universal recommendation score.
- Keep outcomes distinct. Page citation, content-brand recommendation, personal-name mention and personal recommendation are different observations.
- Check the collection interface. Labs #03 and #04 use OpenAI Responses API + web search. They do not represent consumer ChatGPT Search performance.
Labs #01–#05 link to methods and public data. Lab #05 publishes reviewed aggregates. Labs #06 and #07 include methods and evidence links on their report pages. Private originals and account evidence remain outside this public repository. Use GitHub’s Copy permalink to cite a fixed revision, and include the answer or observation ID.
New questions and experiments will be added as learning continues. Earlier methods, corrections and limitations remain visible in the record.