GEO, AEO, AIO, and AI SEO: what are we actually optimizing?

This table is GeoAeoAle’s operating agreement, not an industry law. In every report, expand the label and name the event that was actually observed.

Section sources:[1] Google Search Central

1) Does the URL fail to return meaningful HTML or return an error? Start with crawlability, status codes, robots, sitemap, and internal links. 2) Is HTML available but unclear about its subject and question? Rewrite the heading, direct answer, entity definitions, and date. 3) Is the claim clear but unverifiable? Put primary evidence beside it, with data period, method, and limitation. 4) Are those sound but the answer omits the page? That is a surface observation; do not call it proof of bad content. Record question, mode, language, country, and repeat. 5) Did a citation appear? Check whether it supports the nearby claim and whether it produced a referral. Site owners can improve the first three layers, but retrieval and citation selection remain platform decisions.

Section sources:[4] OpenAI Developers

A team publishes “How to check robots.txt.” It states: robots.txt helps compliant crawlers understand rules but does not protect secrets. The HTML defines the term, links to Google documentation, and gives a curl check; the ledger marks the claim observed, while advice about private routes is an inference. CI checks 200, canonical, sitemap, and no mandatory JavaScript. The editor then asks the same question on selected AI surfaces and saves answers. A bot visit is recorded as fetch; a displayed link as citation. The events are not merged into “GEO +20%.”

Section sources:[1] Google Search Central

A page used to begin with a long history, with its answer five screens down. The editor moves the conclusion up without changing facts and adds a check date. The comparison must report page version, dates, prompt set, repeats, and a control URL. More mentions after the change are an observation. Calling it an effect requires stable repeats and a control; the index, interface, seasonality, or prompt may have changed. Corrections record substantive changes only; modified date is not refreshed merely to look active.

Section sources:[4] OpenAI Developers

Create a URL register and an intent-based panel: definition, comparison, selection, freshness, and troubleshooting. For each run store exact prompt, locale, country, mode, timestamp, full answer, and URLs. Separate server fetches, indexed URLs from webmaster tools, mentions, citations, referrals, and conversions. For every rate show numerator, denominator, and period: 8/40 answers with a relevant link. A User-Agent alone does not verify a bot, and a bot visit does not prove citation. The dashboard needs observed, inference, and hypothesis fields so a hypothesis cannot look like fact.

Section sources:[3] Google Search Central

Do not create hundreds of pages by swapping a city or keyword; that does not deepen knowledge. Do not hide core text behind JavaScript, add JSON-LD absent from the page, or call a special “AI tag” mandatory without an official source. A link list is not evidence: each link must support a specific claim. Do not compare offline chat with web-search mode, publish private client slices, or call correlation causation. Never promise ranking, retrieval, citation, or traffic.

Section sources:[3] Google Search Central

Google confirms search fundamentals for its AI features, but does not disclose a universal source-selection formula. OpenAI documents crawler roles, but one bot visit does not show whether a URL was used in an answer. Every test is bounded by surface, date, language, country, and prompt set. GeoAeoAle therefore publishes reproducible observations, separates inferences from hypotheses, and keeps limitations beside every number.

Section sources:[3] Google Search Central

Before any change, record the question, surface, language, country, and current URL version. Check access next: final status, HTML without JavaScript, canonical, robots, and sitemap. Then isolate one supportable passage and link it to a source. Run a control in the same mode, save the full answer, and record fetch, mention, citation, and referral separately. If page structure changes, do not change the copy and prompt set at the same time. Repeat after a predefined interval and compare absolute counts as well as rates. This protocol does not reveal a hidden algorithm, but it makes an editorial decision reproducible and shows which layer changed.

Section sources:[1] Google Search Central[4] OpenAI Developers