If an AI search result you've been keeping an eye on suddenly changed its answer, you probably weren't imagining it. Google has been routing some AI Mode answers through a lighter, faster version of its Gemini model — and faster models pick businesses differently than slower ones do.

The short version: fast models reward businesses whose basic facts are consistent and easy to find, and skip businesses whose details are scattered or contradictory. You can't control which model answers your customer's question. You can control how much work that model has to do to find you.

What actually changed

Google doesn't use a single model to answer everything. It routes questions between several, depending on how complex the query is and how much capacity it has. Reports suggest a portion of AI Mode answers are now being served by a "Flash" model — an industry term for a lighter, quicker version of a model, built for speed and volume rather than deep reasoning.

That's not a downgrade for users. Most questions don't need heavy reasoning. But it does change the maths for anyone hoping to be recommended.

Why speed changes who gets recommended

A slower, heavier model has room to chase down a fuzzy answer. If your website says one thing and your Google Business Profile says another, it can weigh both, notice which one looks more recent, and reconcile them.

A fast model doesn't do that. It grabs what's easy to find, cross-checks the most repeated version, and moves on. Businesses with clean, consistent, obvious facts get pulled into the answer. Businesses whose details are inconsistent get skipped — not because anything is wrong with them, but because confirming their details costs time the model isn't spending.

Here's what that looks like in practice.

A plumber's homepage says "24/7 emergency callouts across Leeds." Their Google Business Profile lists hours as Mon–Fri, 9–5. Their Yelp listing still shows the suburb they moved out of two years ago.

Someone asks an AI assistant for an emergency plumber at 11pm. A careful model might work out that the 24/7 claim is the current one. A fast model just recommends the plumber whose "24/7 emergency callouts" appears identically in four different places.

Nothing about the first plumber's business is worse. They just made the machine hesitate, and hesitation is expensive when the model is optimised for speed.

The Five Facts: your 30 minutes this week

This is the fix, and it doesn't require a developer.

Write down these five things, once, in plain words:

1. Your business name — exactly as you want it said out loud 2. What you do — one sentence, no marketing language 3. The area you actually serve — named towns or suburbs, not "the surrounding region" 4. Your hours — including emergency or after-hours availability, stated explicitly 5. Your starting price or price range — even a rough band beats saying nothing

Now make those five facts identical everywhere they appear:

Same wording, not just the same meaning. "Open 24/7 for emergencies" and "available around the clock" mean the same thing to a person. To a fast model doing a quick cross-check, they're two different strings that don't confirm each other.

That's the whole exercise. You're not adding anything clever. You're removing every reason for a fast model to hesitate.

For agencies: this is a repeatable first deliverable

If you manage several clients, the Five Facts makes a clean, low-effort onboarding task. Build it as a one-page sheet per client, get it signed off, then push the exact same wording out to the website, the Google Business Profile, and the main directories.

It's cheap to do, easy to explain to a non-technical client, and it produces visible results in AI answers faster than most content work. It also gives you a baseline: when a client asks why they're not being recommended, you can point at a document and say what's already been standardised and what hasn't.

Re-check it quarterly. Directory listings drift, staff update hours in one place and not another, and prices change without anyone remembering the About page.

The bigger point

Model changes will keep happening, usually without announcement. Google will route differently next quarter. OpenAI will ship something new. Chasing each change is a losing game — by the time you've adapted, the thing you adapted to has moved.

What survives every swap is the same: facts that are consistent, specific, and repeated in the places these systems already read. A heavy model rewards that. A fast model rewards it more.

Get the Five Facts right and you stop caring which model answers the question.

Where to start

Pick the fact most likely to be wrong. For most local businesses, that's hours or service area — the two things that change quietly and rarely get updated everywhere.

Search your own business name in an AI assistant and ask it what your hours are and where you operate. If the answer doesn't match what you'd say yourself, you've found your first fix.

If you'd rather see all five checked at once — including which outdated listing an assistant is pulling from — a SimpleAIO audit shows you in plain English, with the fixes ranked by what to do first.