Reports this month suggest Google has started serving some AI Mode answers with Gemini 3.7 Flash — a lighter, faster model in the Gemini family. Google routes queries across multiple models depending on complexity and load, so the same question can be answered by different systems on different days.

For agencies, this is less a technical story than a client-communication and pricing story.

What changes for your clients

Faster models do less reconciliation work. A heavier model will chase down a contradiction between a client's website and their Google Business Profile. A Flash-class model grabs the most easily confirmed version and moves on.

So the clients who suffer aren't the ones with bad websites. They're the ones with inconsistent facts — a service area that changed two years ago and never got updated on three directories, hours that differ between the site footer and the GBP listing, a service name written four different ways.

Those clients were previously getting the benefit of the doubt. Under a faster model, they don't.

The conversation to have this month

You will get a call from a client saying "we disappeared from AI answers." Get ahead of it.

Send a short note now, before the call: Google is rotating models behind AI Mode, answers will fluctuate week to week, single-day checks are noise, and here's what we monitor instead. Clients who hear this from you first treat volatility as expected. Clients who discover it themselves treat it as your failure.

This note costs you twenty minutes and saves a retainer.

How to package it

Model churn is the strongest argument you have against one-off AEO/GEO audits — the deliverable expires. Three offers that work:

Fact Alignment sprint (one-off, £900–£2,000 / $1,200–$2,500). Audit the client's five core facts — name, one-line service description, service area, hours, price range — and make them word-for-word identical across the website, Google Business Profile, and the top directories in their vertical. Fixed scope, junior-deliverable, high perceived value.

AI visibility monitoring (retainer, £250–£600 / $300–$750 per month). A defined prompt set checked monthly, tracked over time, reported with a short written read. The value is the trend line and the interpretation, not the raw numbers.

Model-shift response (retained hours). When a rollout moves results, you already have the baseline to say what actually changed. Clients pay for that certainty.

What it means for your own shop

Two things worth noting.

First, margin. Fact alignment is checklist work. It can be delivered by a coordinator against a template, which makes it one of the healthier-margin services in the AEO/GEO category.

Second, defensibility. Anyone can run a client's name through ChatGPT once. What competitors can't replicate is twelve months of your monitoring history. Start baselines now on every account, even the ones not paying for it yet — the data becomes the pitch.

If you'd like a fast baseline across your book, SimpleAIO audits each client's AI visibility and returns plain-English fixes you can hand straight to your delivery team — or attach to the proposal.