How to Win in Google’s AI Mode Era: SEO, Ads and Brand Strategy

Search changed. Did your strategy work?

Google’s AI Mode does not return a page of links. It returns an answer, then lets the person ask the next question, and the next, until they have decided.

That is a different competition from ranking. In a list, position eight still exists. In a composed answer there are two or three businesses named and everyone else is absent, and the person can keep interrogating the answer without ever seeing a results page.

This post covers both sides of what that changes, because most coverage only handles one. The organic side is about being readable and verifiable enough to be named. The paid side is about feeding Google’s AI-driven campaign types signals clean enough to work with. Most businesses are neglecting one of the two, and usually it is the second.

What AI Mode Actually Is

Worth taking from Google directly rather than from interpretation.

Google introduced AI Mode at its I/O conference in May 2025, describing it as “our most powerful AI search, with more advanced reasoning and multimodality, and the ability to go deeper through follow-up questions and helpful links to the web.”

Two words in that sentence matter more than the rest. Reasoning, because the system is assembling rather than retrieving, which means it decides what it can state confidently. And follow-up, because the conversation continues after the first answer, and most businesses only ever consider the first question.

On scale, this is where a lot of published advice is out of date, including the earlier version of this page. In October 2025 Google announced that AI Mode had expanded by “more than 35 new languages and over 40 new countries and territories”, reaching “over 200 countries and territories total”.

If you read somewhere that AI Mode is a limited pilot in a handful of markets, that was true for about four months in 2025. It is not a pilot. It is the default for a very large share of the planet.

Why old SEO and PPC stopped being enough

Neither is wrong. Both are incomplete, and they are incomplete in the same way.

On the organic side, the work that earns a ranking and the work that earns a citation overlap heavily but are not identical. A page can rank well and never be quoted, because ranking tolerates ambiguity and quoting does not. A system deciding whether to name your business has to be able to state something about you without hedging, and anything it had to interpret is a risk it can avoid by using a clearer source.

On the paid side, the change is less discussed and arguably larger. Google’s campaign types increasingly make the bidding, placement and creative-assembly decisions themselves. What you control is the quality of what you feed them. Campaigns with broken conversion tracking and no first-party audience data are asking an automated system to optimize toward a signal that is wrong, and it will do exactly that, efficiently.

The connecting idea: on both sides, you stopped being the operator and became the source of truth. What the machine knows about your business, and how confidently it knows it, now decides the outcome on both.

The AI Mode playbook

Six steps, in the order that works. Organic first, because the paid side inherits its data quality from the same foundations.

1. Answer the question, then keep answering

Structure pages around the questions customers actually ask, with the answer in the first two sentences under a heading phrased the way they would ask it. Then handle the follow-up in the same page, because AI Mode’s whole design is the follow-up. “How much does it cost” is followed by “what affects the price” and “why are you more expensive than the other quote”. A page that answers only the first question exits the conversation early.

2. Make your facts machine-readable

Structured data on your key pages, generated from your own fields rather than pasted, so your facts are stated rather than inferred. Hours, services, prices where you publish them, and identity. This is the difference between a system reading your business and guessing at it.

3. Make them agree everywhere

Your site, your Business Profile, your listings, your marketplace pages. Where they agree, a system has a fact. Where they conflict, it has a reason to use somebody else. For local businesses this is usually the largest single gap, and we covered how AI reads a profile specifically in our guide to how AI reads your Google Business Profile.

4. Fix your conversion tracking before you touch bids

This is the paid-side foundation and it is the step most often skipped. If your conversion actions are miscounted, double-counted, or measuring form loads rather than qualified enquiries, every automated decision downstream is optimizing toward the wrong thing. Automation amplifies whatever you point it at.

5. Give the algorithm first-party data

Customer lists, purchase data, offline conversion imports where you have them. Automated campaign types perform in proportion to the quality of the audience signals they receive, and first-party data is the one input your competitors cannot copy. This is also the part that compounds, because your data improves as you use it and theirs does not.

6. Write creative a machine can classify

Automated systems assemble your assets rather than running them as you wrote them, which means clarity beats cleverness. Headlines that state what the thing is, images where the subject is obvious, and asset sets covering your actual services rather than five variations of one message. Ambiguous creative gets assembled ambiguously.

Google Ads inside AI: what “training the algorithm” actually means

The phrase gets used loosely, so here is the concrete version.

You are not training a model in any technical sense. You are supplying the inputs a decisioning system uses, and the quality of those inputs sets the ceiling on what it can achieve. Three inputs do most of the work.

Conversion data is the objective: The system optimizes toward whatever you told it success looks like. If that definition is loose, it will find you a great deal of loose success. Counting every form submission as a conversion when a third are spam teaches it to find more spam. Getting this right is unglamorous, and it is worth more than any bid adjustment.

Audience signals are the starting point: First-party lists give the system a shape to look for. Without them it explores from a cold start, spends more finding the pattern, and takes longer to stabilise.

Creative is the vocabulary: Assets are combined dynamically, so the system needs enough distinct, clearly-labelled material to assemble something coherent for each context. Thin asset sets produce repetitive combinations regardless of budget.

There is a fourth input people forget, and it is your website. Automated campaigns read landing pages to decide what to say about you and where to send people, which means the same clarity work that earns an organic mention also improves what the ad system understands. That is the practical argument for running the two disciplines together rather than in separate agencies with separate reports: they are reading the same pages and rewarding the same qualities.

The uncomfortable implication for most advertisers: if performance is poor, the lever is usually not inside the ad platform’s settings. It is in the measurement and data layer feeding it, which is nobody’s favourite project and the only one that reliably moves the number.

The wider shift in paid, including what happened when ChatGPT started selling ads, is in our guide to the future of AI advertising. If you want that side run properly, that is our PPC work.

Test yourself in AI Mode, including the follow-up

Fifteen minutes, monthly, and it measures the outcome rather than a proxy for it.

Ask your category question: The one a customer asks before choosing, not your brand name. Note whether you are named.

Then ask the follow-up: This is the part almost nobody tests, and it is where AI Mode differs from every other surface. “Which of those is best for X.” “Which is cheapest.” “Which one handles emergencies.” The businesses that survive three follow-ups are the ones with enough specific, verifiable detail published to keep being relevant as the question narrows.

Then ask an adversarial one: “Any downsides to using them.” A system with nothing to say will say nothing, which is fine. A system repeating an old review or a stale fact tells you exactly what to fix.

Write down who else appears: That list is your competitive set as the machine understands it, which is often not the one you assumed.

Two things worth knowing before you read too much into any single run. Answers vary between sessions and between people, so a business that appears once and not twice has not necessarily changed anything; the pattern across a few months is the signal, not any individual result. And your own account history influences what you see, which is why the test only works signed out. Neither caveat makes the exercise less useful. They make it a trend to watch rather than a score to react to.

Do it monthly and keep the notes. Four data points across a quarter tell you more about your AI visibility than any dashboard currently sold. The broader discipline behind it is answer engine optimization, and the comparison with how ChatGPT handles the same job is in our post on ChatGPT search versus Google.

Where to start this quarter

Three moves, and the first two are free.

Run the test above: Question, follow-up, adversarial follow-up. Write down who gets named.

Audit your conversion tracking: Not your campaigns. The definitions underneath them. Most accounts have at least one conversion action counting something that is not a customer.

Fix the facts the test exposed: Usually a handful of questions nobody has answered directly, and details that disagree between your site and your profiles.

If you would rather have both sides run together, our SEO and paid teams work from the same audit rather than separate ones, and contact us starts with what AI Mode currently says about your business.