Microsoft Told Retailers How AI Picks Winners. Here's the Part That Applies If You Don't Sell Products.

Microsoft Advertising published a guide called "From Discovery to Influence: A Guide to GEO". It is about sixteen pages, it is free, and it is genuinely one of the better explanations of how AI-driven discovery works.

One aside on the name, because it is a small lesson in itself: Microsoft markets the document as a guide to "AEO and GEO", and the download even sits at aka.ms/ads/AeoGeo, but open it and the title is "A Guide to GEO", with AEO never actually defined anywhere inside. We wrote separately about that acronym muddle. Here we care about the substance, which is good.

That substance is written for retailers. Open it expecting advice for your accountancy practice or your physiotherapy clinic and you will hit product feeds, GTINs, SKUs, stock levels and shopping schema within a couple of pages. A note on where it comes from, too: this is Microsoft's advertising team, not its search engineering team, writing for merchants who buy shopping ads. That does not make it wrong. It does mean it is a vendor playbook with a commercial interest in retailers investing in feeds, so read it as informed advocacy rather than neutral documentation.

Strip that away and three ideas underneath are worth more to a service business than most of what is being written about AI visibility right now.

Idea one: there are three routes to information about you

Microsoft's core model is that AI systems draw on three separate pathways, and you need to be present in all of them. For a retailer those are crawled web data, product feeds, and the live website an agent visits. For a service business the middle one looks like it does not exist. It does. It just is not called a feed.

Crawled data is the same for everyone: your website, and everything written about you elsewhere. This shapes whether an AI has any coherent idea of who you are and what category you belong to.

Your structured data is your feed. A retailer pushes price, availability and specs through a product feed. You push the equivalent through schema markup, your Google Business Profile and consistent listings: your services, your location, your hours, your credentials, your named people. It is the same job, the same machine-readable precision, delivered by different plumbing. Most service businesses have almost nothing here, which is the single biggest gap this guide exposes.

Live site data is where it gets uncomfortable. Microsoft's point is that AI does not just read your site, it acts on it, and a broken live experience means a lost sale. For a retailer that means checkout. For you it means the booking form, the enquiry form, the phone number, the availability calendar. If someone, or something, arrives ready to act and the path is broken, the recommendation was wasted.

Idea two: completeness beats cleverness

The line from Microsoft's own announcement is blunt: completeness beats cleverness, and entries with more fields filled in do better. Alongside it sits the observation that most brands already have the data AI needs, it is just buried.

That is the most useful sentence in the whole document for a service business, because it reframes the work. You probably are not missing information. Your qualifications, your prices, your service areas, your opening hours, your specialisms, the names and credentials of your team all exist. They are in your head, in a PDF, in an Instagram bio, in a booking system, in a photo of a price list.

The job is not to invent something clever. It is to get what you already know into plain, structured, machine-readable text on your own site. That is unglamorous, and it is most of the win.

Idea three: the clarity ladder, translated

Microsoft's neatest illustration is a before and after, using a rain jacket. The plain SEO version is the bare keyword: "waterproof rain jacket". The GEO version is the same product carrying its justification: "best-rated waterproof jacket by an outdoor magazine, no-hassle returns for 180 days, three-year warranty, 4.8 stars". Keyword in, credibility out.

We find it clearer as three rungs than two, because the middle step is where most service businesses actually get stuck:

1. The keyword. What you would once have optimised for. "Accountant Manchester."
2. The specifics. Enough detail that a machine, and a buyer, understands exactly what you do and who for. "Chartered accountancy firm in Manchester, year-end accounts and tax planning for owner-managed businesses turning over £500k to £5m, fixed monthly fees."
3. The proof. The credibility that makes an AI willing to recommend you over the firm next door. "ICAEW chartered, 4.9 stars across 120 Google reviews, twenty years in practice, free switching service."

Most service websites stop at the first rung. The businesses AI recommends have climbed all three. Nothing about that requires a product catalogue.

What to ignore

If you sell services, you can skip the parts of the guide about product feeds and feed management, GTIN, SKU and variant fields, inventory and stock signals, Offer and Product schema, and anything about shopping surfaces or checkout agents. None of it applies to you, and there is no service-business equivalent worth chasing.

Skip, too, the conclusion some people are drawing from this document: that you need to publish dozens of articles at scale to compete. That is not in Microsoft's guide. It is a sales pitch attached to it, usually by someone selling content bundles. As we covered when Google documented query fan-out, mass-producing pages to chase AI answers is the behaviour Google's scaled content abuse policy exists to catch. Coverage of your customer's real decision is the goal. Volume is not.

What to actually do

Three moves, in order. None of them is exotic, which is rather the point.

• Get what you already know into structured, readable form. Add schema for your organisation, your services, your location and your named people. Fill in your Google Business Profile completely rather than partially. Write your credentials, prices and service areas as plain text on your own pages, not as images or PDFs.
• Climb the ladder. Take your main service page from a keyword, to a specific description of what it is and who it is for, to a version that also carries the proof: qualifications, reviews, guarantees, years in practice.
• Check the last mile. Open your own site on a phone and try to book yourself. If the form is broken, the number is not tappable or the calendar is empty, fix that before anything else. Being recommended and being unbookable is the most expensive failure available.

The retail specifics in Microsoft's guide will not help you. The model underneath it will, and it says the same thing every credible source now says: be readable, be complete, be credible, and be able to take the booking.