How to Get Your Business Recommended by ChatGPT and AI Search

Ask ChatGPT "who should I hire to do my bookkeeping in Manchester?" and it will answer. Not with ten blue links — with two or three named firms, a sentence on each, and a reason to pick one. Ask Perplexity the same thing and you get a shortlist with citations. Your prospective customer never sees a search results page, never scrolls, and never discovers you exist unless the model happened to name you.

Traditional SEO is about being ranked. AI search is about being recommended. They are related but not the same, and the tactics that get you cited in an AI answer are narrower and cheaper than a full SEO programme. Here is exactly what to do, in the order that matters.

Why AI search now decides who gets shortlisted

The behavioural change is simple to describe. A buyer used to search, skim five results, open three tabs and form a shortlist themselves. Now a meaningful share of that first step happens inside an AI assistant, and the assistant hands over a shortlist that is already down to two or three names. You are either on it or you are not in the conversation.

This cuts both ways for a small business. The bad news: there is no page two to be found on. The good news: models do not rank by domain authority the way a search engine does, so a well-documented five-person firm can be named alongside a national chain. Size is less of a moat here than it was.

One owner put it the way a lot of people feel: "Every article says 'use AI' but nobody says HOW." Fair. So here is the how, with no vague platitudes.

How AI assistants actually pick which businesses to name

It is worth being precise about the mechanism. When an assistant answers a question about local or specialist providers, it is usually doing one of two things.

Retrieving live results. Most assistants run a live web search behind the scenes, pull the top handful of pages and write an answer from them. Classic search visibility therefore still matters — but the pages that get quoted are the ones that answer the question directly and plainly.

Answering from training data. Where the model answers without searching, it draws on what it absorbed during training: directories, review sites, news coverage, forum threads, trade listings, your own site. Slower-moving and harder to influence — but where long-standing reputation pays off.

The practical consequence: your job is not to "rank" but to be the easiest thing to quote on the question your customer is asking. That means clear claims, specific facts, and corroboration on sources the model already trusts.

Step 1: Find out what AI already says about you

Do not buy a tool for this. Twenty minutes and a free account will do.

Write down the ten questions a good-fit customer would genuinely type. Not "best accountant" — the real ones, with the qualifier attached: "affordable accountant for a limited company in Leeds", "who can do R&D tax credits for a small software firm", "best physiotherapist near Camden for running injuries". Then run each one through ChatGPT, Perplexity, Google's AI answers and Copilot, and record three things in a spreadsheet:

• Were you named? Yes or no, per question, per assistant.
• Who was named instead? These are your real competitors for this channel, and they are often not the competitors you think.
• What did it cite? Click through. If the model cited a directory, a review site or a trade listing, that source is now a priority for you.

Repeat this quarterly. It is the only honest baseline you will get.

Step 2: Fix the sources AI reads before your website

This is the step most owners skip, and it is usually the one with the biggest return. The models lean heavily on third-party sources because those look more like corroboration than a business describing itself.

Your Google Business Profile. Complete every field. Categories, service list, service area, opening hours, attributes, and at least a dozen photos. The description should state plainly who you serve and what you do not do. "We do bookkeeping and VAT returns for UK limited companies with 1–20 staff. We do not do personal tax returns." That specificity is exactly what makes a model comfortable naming you for a matching query — and comfortable leaving you out of a mismatched one, which is also a win.

Reviews with substance. A five-star review that says "Great service!" is worthless to a language model. One that says "They sorted out three years of messy Xero data for our 12-person agency in about two weeks" is a quotable fact. When you ask for reviews, ask specifically: what problem did we solve, and roughly how long did it take? You are not gaming anything — you are making genuine outcomes legible.

Directories and trade bodies. Your professional association listing, Companies House data, Crunchbase, sector directories, the local chamber. These are heavily represented in training data and retrieval alike. Make sure your name, address and description are identical everywhere. Inconsistency makes you look like three half-real businesses instead of one solid one.

Step 3: Rewrite your site so an AI can quote it

Most small business websites are written to sound impressive. AI search rewards sites written to be useful and specific. Three changes:

Answer the question in the first two sentences. Under each heading, lead with the direct answer, then explain. Models extract the opening lines of a relevant section far more often than the middle of a long paragraph. A page that opens "Our VAT return service costs £95 per quarter for companies turning over under £500,000" is quotable. One that opens "We pride ourselves on a bespoke, client-centric approach" is not.

Put the boring facts in plain text. Prices or price ranges, service areas, team size, typical turnaround, what you do not do, sectors you specialise in, languages you work in. Owners often bury these or leave them off entirely. If a model cannot find your price band, it will name a competitor whose price band it can find.

Add a genuine FAQ and schema markup. Eight to twelve real customer questions with direct answers, marked up with FAQPage schema, plus LocalBusiness or ProfessionalService schema on your homepage covering name, address, phone, opening hours, service area and price range. Schema is not a ranking trick; it is a way of stating facts unambiguously so a machine does not have to infer them. Our guide on how to use AI for SEO in a small business covers the drafting side of this in more detail.

Step 4: Get mentioned where AI looks for opinions

When someone asks an assistant for a recommendation rather than a fact, the model reaches for places where humans give opinions: Reddit threads, industry forums, YouTube, podcasts, local news, trade press, and roundup articles.

You cannot spam these, and attempts tend to backfire publicly. What works is slower and more ordinary:

• Answer questions in your sector's forums and subreddits as yourself, with your business in your profile, giving real advice without pitching. Ten genuinely helpful answers over a quarter is a realistic target.
• Get onto one relevant "best X in Y" roundup per quarter. Most are compiled by small publishers who accept submissions. These roundups are retrieval gold because they are structurally a list of named businesses.
• Do one podcast or local press piece a year. A named interview ties your business to a topic in a way that survives for years.
• Publish something with a number in it. Original data — even a small survey of 40 customers — gets cited far beyond its size, because models favour sources that state specific figures.

Step 5: Measure it without buying another tool

There is already a market of AI-visibility tools charging €80–€300 a month. Most SMBs do not need one yet. Three free signals will tell you whether this is working:

Your quarterly question audit from Step 1. Track the share of your ten questions where you are named. Going from 1/10 to 4/10 over two quarters is a real, measurable result.

Referral traffic from AI assistants. In your analytics, check for referrers from chatgpt.com, perplexity.ai, copilot.microsoft.com and similar. Volume is usually low but conversion rates are typically strong, because the visitor arrives pre-qualified by the recommendation.

Ask new enquiries how they found you. One line on your enquiry form. "ChatGPT suggested you" is now a real answer people give, and it is the cleanest evidence you will get.

Revisit after two quarters, not two weeks — reacting to a fortnight of noise is how people waste money here. The same discipline applies to any AI initiative, and our walkthrough on how to calculate the ROI of AI implementation sets out how to judge one properly.

What not to bother with

A quick list, because money is currently being spent badly here.

Do not pay an agency €2,000 a month for "GEO" or "AEO" services until you have done Steps 1 to 3 yourself. Those steps are most of the available gain, and they cost time rather than budget.

Do not stuff your site with invisible instructions aimed at AI models. It does not work reliably, and where it does it is the kind of thing that gets penalised the moment it is detected.

Do not abandon conventional SEO. Retrieval-based assistants search the web to answer questions. Being findable by a search engine is still a precondition for being quoted by an assistant — the two reinforce each other rather than competing.

Do not chase every new assistant. Cover ChatGPT, Google's AI answers, Perplexity and Copilot. That is the great majority of the volume. Chasing the rest is the "six AI subscriptions, no coherent workflow" trap in a different costume.

The bottom line

Getting recommended by AI search is not a new discipline requiring new software. It is the old discipline of being specific, well-documented and genuinely well-regarded — applied to a reader that cannot infer, cannot phone you to check, and will name someone else if your facts are missing.

Start with the twenty-minute audit. Fix your Google Business Profile and your two most important directory listings. Rewrite your three main service pages to answer questions directly and state the boring facts in plain text. Then let it run for a quarter and re-audit. For most small businesses that is a few days of work against a channel that is only growing.

And keep it in proportion: this is one channel, not a strategy. If you have not yet stepped back to work out where AI genuinely changes your business rather than just your marketing, our guide on how to create an AI strategy for a small business is the better place to start.