Published
Updated
What this reference is
Hartzer.it.com is a technical reference on AI search visibility: how generative and answer engines decide which web pages to retrieve, which to cite, and how to attribute them. It covers the surfaces themselves — the products that write an answer and attach links — and the techniques publishers are told to adopt in order to appear in those answers.
It exists because the gap between what the platforms document and what the market sells is unusually wide in this subject, and unusually easy to close. Google, OpenAI, Microsoft and Perplexity all publish documentation. Independent researchers have run replicable studies. Most of the advice circulating about generative engine optimization cites neither, and a good deal of it contradicts both. A page that reads the documentation, reads the studies, and then says plainly which of the two the advice matches is not a difficult page to write. It is simply one almost nobody writes.
Who writes it
Every page here is written by Bill Hartzer, who has worked in organic search since 1996. That span is the relevant qualification, and it is relevant for a specific reason rather than as a length-of-service claim: the mechanics that decide whether an AI answer cites a page are, in large part, the mechanics that decided whether a results page ranked one. Retrieval still runs on an index. Eligibility still runs on crawling and indexing. Passage-level matching, entity resolution and answer-shaped writing all have a decade or more of prior art in featured snippets and passage ranking, and the people who worked through those transitions recognise which parts of the current wave are new and which parts have simply been renamed.
First-hand experience matters here in a second way. Claims about AI search are usually argued from screenshots. Screenshots are personalised, geographically variable, and frequently irreproducible an hour later. Knowing how to construct a test that survives that — controlling the query set, the locale, the account state and the time window — is the difference between measurement and anecdote, and it is a skill that came from a long run of doing the same thing to search results.
Bill's professional profiles are hartzer.com and billhartzer.net. His consulting and professional work is not conducted here and is not described here.
The evidence verdict
Every entity page and every guide leads with a coloured verdict before a word of prose. There are three states and they are used strictly.
- Documented — the platform states this in its own documentation, or independent replication supports it. The page cites the source.
- Observed — practitioners see it consistently, the platform has not confirmed it, and it may change without notice or explanation.
- Unsupported — the claim is widely repeated and there is no evidence for it. The page says so, and then explains why the belief persists, because the reason a false claim spreads is usually more useful than the correction.
The third verdict is the one that makes the other two worth reading. A reference that only ever agrees with the consensus adds nothing to the consensus. A substantial share of the pages here land on Unsupported, and that is a finding about the subject, not an editorial posture.
How a page gets written
Each page starts from primary sources: platform documentation, engineering and product blogs, published research, and the reporting surfaces the platforms actually expose to publishers. Secondary commentary is read but is not treated as evidence. Where a study is cited, the page names who ran it and what they measured, because sample construction is where most AI search studies fail — a citation-share figure taken from a thousand self-selected prompts is not a measurement of anything a reader can act on.
Dates are load-bearing in this subject and are treated as such. A statement that was true of AI Overviews in mid-2024 may be false now; a documentation page updated last December supersedes a conference talk from the spring. Where a claim depends on a date, the page carries the date.
What this site does not do
It sells nothing. There are no engagements, no rates, no proposals and no products. There is no affiliate link anywhere on the site and no tool is ranked, scored or recommended. Tools are named where naming them is the honest way to explain what a category of tool measures, and the page then explains where that measurement misleads.
It runs no analytics and sets no cookies of its own. It carries no advertising. It publishes no sponsored content and accepts none.
Several adjacent subjects are deliberately out of scope because they are covered properly elsewhere and duplicating them here would serve nobody. Professional services, industry commentary, and the author's own publication and speaking record all live on their own sites, not on this one.
Corrections
Corrections are welcome and are the most valuable message this site receives. If a page states something the documentation does not support, or a study is characterised inaccurately, or a platform has changed its behaviour since a page was written, the contact page explains how to send that. Corrections are made to the page itself, and the modification date on the page moves when they are.