What You Can Influence in Google's AI

There is no such thing as "Google's AI" in the singular, and that is the first reason most advice on this subject is useless. When someone says they want to be recommended by Google's AI, they are usually thinking of one specific thing they saw — an AI Overview at the top of a results page, or an answer in AI Mode — and generalizing it into a single system with a single set of rules.

Google is not one surface. It is several, they retrieve and present information differently, and work that helps on one can be irrelevant on another. Being precise about which one you mean is the difference between a plan and a wish.

The surfaces, and why the distinction matters

Organic web results. The classic list of links. Drawn from Google's web index and driven by the crawled, indexed content of pages.

Local results and the map pack. The block of businesses with a map. Drawn from the local index and driven overwhelmingly by your Google Business Profile, the searcher's proximity, and prominence. Your website matters here, but less than most people assume.

Google Maps. The Maps app and maps.google.com. Related to local results, and frequently ordered differently for the same words, because the intent Maps infers is not the intent a web search infers.

AI Overviews. The generated summary that sometimes appears above results, grounded in web results and linking out to sources.

AI Mode. A separate conversational experience — longer, more decomposed questions, and follow-ups.

Google Business Profile. Not a ranking surface at all. It is a data source that feeds local results, Maps and knowledge panels. It is also the single asset a local business most directly controls.

Paid advertising. An auction, entirely separate from everything above.

Improving your Business Profile can move local results and Maps and do nothing whatsoever for AI Overviews. A page that gets cited in an AI Overview may sit on page three organically. Treating these as one system is why so much local marketing advice produces nothing measurable.

What Google actually documents

Worth reading in the original rather than through somebody's summary:

• Search Essentials — the technical, spam and quality requirements.
• Creating helpful, reliable, people-first content — the content standard.
• Improve your local ranking on Google — which names the local factors as relevance, distance and prominence.
• Top ways to ensure your content performs well in AI experiences — which states that there is no special markup or setting that opts a page into AI Overviews, and that the ordinary crawling and indexing rules apply.

That last point deserves emphasis, because it contradicts a great deal of what gets sold. There is no AI schema. There is no AI Overviews submission form. There is no setting to switch on.

What remains undocumented

Google does not publish:

• How relevance, distance and prominence are weighted, or how that varies by query and industry.
• Which sources a particular AI Overview drew on beyond the links it chose to show.
• Why one business appears and a comparable one down the road does not.
• How often any of it is recomputed.
• Whether a given answer was personalized, and by what.

So when you see a ranked list of "AI recommendation factors" with percentages attached, those percentages were inferred, not obtained. The inference may still be useful. It should be labeled as inference, and it usually is not.

What a local business genuinely controls

The list is short, which is the point.

• The accuracy and completeness of your Business Profile — categories, service areas, hours, services, attributes, photos, and whether all of it is currently true.
• Consistency of your name, address and phone number everywhere they appear.
• What your website says about what you do, where you do it and who you serve — in text a machine can read, not baked into an image.
• Whether the site is crawlable, indexable and quick enough to be fetched.
• Structured data that accurately describes what is visible on the page.
• Whether real customers leave reviews, and whether you answer them.
• Whether third parties — directories, licensing bodies, trade associations, local press — carry correct information about you.

That is most of it. It is unglamorous, and it is also the only part of this system where your effort maps onto something you can check.

What Google controls

Everything else. Whether you appear in any given answer. Which sources a generated answer draws on. Whether an AI Overview appears for that question at all. How much of the answer is visible before anyone clicks. Whether the surface still exists in this form next year.

No quantity of schema, reviews or published content converts into a guarantee of inclusion. I have not found a mechanism that reliably forces it, and I would be wary of anyone who says they have.

How to test what is actually happening

The method matters more than any single result, because a sloppy test produces confident nonsense.

1. Write down the exact question a customer would ask, in their words — not your keyword, their sentence.
2. Record the conditions: date, which surface, device, approximate location, and whether you are signed in.
3. Run it signed out, in a private window. Signed-in results carry your own history, which includes every time you looked up your own business.
4. Screenshot what you get, including any sources cited.
5. Repeat the same question at least three times across several days. Generated answers are not stable and one observation tells you very little.
6. Note factual errors specifically — wrong hours, wrong service area, a competitor's number attached to your name. These are the most fixable findings and the most commercially damaging.
7. Vary the location if you serve more than one area. Distance is an explicit local factor, so a result taken at your office says nothing about a suburb twenty minutes away.

Why one observed answer is not a ranking

This is the part most reporting gets wrong.

A generated answer is a single sample from a system that is location-dependent, time-dependent, sometimes personalized, and not deterministic. Two people asking the identical question from adjacent streets can see different answers, and the same person can see something different an hour later. That is not a malfunction. It is how the surface works.

So "we appeared in an AI Overview" is an observation, not a position. It has no rank, no impression count and no trend unless you sample it repeatedly under recorded conditions. Treat a single screenshot as evidence that inclusion is possible — never as evidence that it is stable.

The useful corollary: if you run the same question ten times over two weeks and never appear, that is a far stronger signal than appearing once. Absence is easier to establish than presence.

How to measure the thing that actually pays

Visibility is a proxy. Customers are not. Track what survives a change in somebody else's product:

• Qualified inquiries — calls, forms and texts from people in your service area asking about work you actually do.
• Booked jobs, and where each came from.
• Cost per acquired customer by channel — not cost per click, not cost per lead.
• Referrer data in your analytics. Assistant traffic arrives under identifiable referrers and is worth segmenting; do not expect it to be large.
• The question you ask on the phone: "How did you hear about us?" Imperfect, and still one of the better instruments available to a small business.

If this work is producing anything, it should eventually show up in one of those. If it shows up in none of them across a couple of quarters, it is not paying for itself, whatever the screenshots say.

Where I come out

I build, host and run websites and advertising for local service companies, so I spend real time on this problem. My honest position is that the work worth doing is the work that was always worth doing: accurate business information everywhere it appears, a site that plainly states what you do and where, real reviews from real customers, and structured data that matches the page.

That work makes a business easier for any retrieval system to find, understand and verify. Whether a particular system then names you is not mine to promise, and I do not promise it. What I will do is say plainly what is testable, test it, and show you what the tests said.