Small businesses can win in AI search, but not by doing a smaller version of what large businesses do. Assistants assemble one answer rather than ranking ten results, so specificity beats scale. The four things worth your time at under 20 staff are entity resolution, a Google Business Profile that is actually complete, 10 to 15 properly answered questions, and consistency maintained for a year.
This page is about how. If you are still deciding whether to bother at all, read is small business AI SEO worth it first, which is the decision page. This one assumes you have decided.
How AI models actually treat a small business
There is a common line that AI “prefers big brands”. It is close enough to feel true and wrong enough to lead you somewhere unhelpful.
What is actually happening is corroboration weighting. When a model assembles an answer, it is more confident about facts that appear in several independent places. Large companies clear that bar without trying, because journalists write about them, directories list them, analysts cover them, and their own employees post about them. A 14-person consultancy in Rotherham may be better at the work and appear in three places total.
That is a corroboration gap, not a size penalty. The distinction matters because one of those can be closed by a small business and the other cannot.
Three consequences follow, and they shape everything below.
Models reduce risk when they are uncertain. Faced with a thin entity, an assistant will recommend a category, a directory or a well-known name rather than a business it cannot verify. You are not being outranked. You are being routed around.
Models reward the source that resolves the question. This is the opening. Large competitors write broad pages for broad terms. Nobody at a 200-person firm is writing the definitive answer to “can a UK charity reclaim VAT on a refurbishment”, and that is exactly the sort of question an assistant needs a source for.
Models notice contradiction more than absence. Being absent from a directory is neutral. Being listed with a five year old address is a signal that something about your entity does not reconcile, and it costs you more than the listing was worth.
The four things worth doing
Ordered by return per hour, not by how interesting they are.
1. Make yourself resolvable. One day of work
This is the whole ballgame for a small business, and most never do it.
Write one canonical description of the business, 40 to 60 words, plain language, specific about what you do and who for. Then make every public surface say it: your website footer and about page, Companies House, your Google Business Profile, LinkedIn, your trade body listing, your invoices, the bio on any directory you are in.
Then check the boring facts agree everywhere. Legal name. Trading name if different. Address. Founding year. Phone number. It is astonishing how often a business has three founding years in circulation.
Then add Organization schema to your site, and confirm your robots.txt is not blocking the AI crawlers. Plenty of small business sites block GPTBot by accident, usually because a plugin default did it.
There is a finish line here. That is rare and you should take advantage of it.
2. Finish your Google Business Profile properly
Not “claim it”. Finish it. Every service listed with its own description, opening hours including bank holidays, the full service area, 20-plus photographs, the questions section answered by you rather than left to the public, and the description written in the same words as your canonical description above.
This feeds Google AI Overviews and Gemini directly and it is a meaningful input to how other assistants understand a local trading entity. It is free, it takes an afternoon, and about half of UK small businesses have it half done.
If you serve customers from a location, the mechanics deserve their own read: AI search for local business.
3. Answer 10 to 15 questions properly
Not 60 pages. Not a blog. Ten to 15 pages, each one the best answer on the UK internet to a question your buyers genuinely ask.
Choose them from your own inbox and phone log rather than from a keyword tool. The questions people ask before they trust you enough to enquire are the questions assistants are being asked, and they are usually more specific and more awkward than anything a tool will surface.
Write each one so the answer is in the first 60 words. Then the detail. Then the honest limitation, which is the part almost nobody includes and the part that most reliably earns a citation, because a model looking for a source that resolves a question prefers one that admits the edges.
Sustained monthly is worth more than 15 published in a fortnight and then silence. Models weight recency, and a site that stopped in March reads as stale by September.
The mechanics are in how to write content AI models actually cite and what content AI prefers to cite.
4. Get corroborated, slowly, for a year
The slow one. Third parties saying you exist and are credible.
For a small business that means: your trade association directory, your local chamber, two or three genuine supplier or partner pages that mention you, any local press that will take a comment, review platforms with real reviews on them, and a founder profile that is actually maintained. It is unglamorous and it is the thing that eventually moves you from “a business that exists” to “a business an assistant will name”.
The difference between link building and citation building is worth understanding before you spend anything here, because the tactics are not the same and the small business budget does not survive doing both.
Three things that only work at scale
Say the uncomfortable thing.
Content volume. Publishing 40 or 60 pages a quarter is a legitimate enterprise strategy. At small scale it produces thin near-duplicates that make a model less confident about what you specialise in. You will be paying to blur your own entity.
Broad category terms. “Accountant UK” is not available to you and pursuing it will absorb the whole budget. “Accountant for UK ecommerce sellers dealing with EU VAT” is available, and is a better lead anyway.
Tracking platforms. They are good tools priced for portfolios. With 20 prompts you can test them by hand in an hour a month, and you will learn more doing it, because you will read the actual answers rather than a number. Buy one when the manual test genuinely becomes the bottleneck. The tool versus agency comparison covers where the line sits.
A realistic 12-month shape
| Months | Focus | Time cost |
|---|---|---|
| 1 | Entity resolution, Google Business Profile, schema, crawler access | 2 to 3 days total |
| 1 to 3 | Baseline: 20 prompts tested across 5 assistants, recorded | Half a day, then an hour a month |
| 2 to 6 | 10 to 15 question pages, one or two a month | 4 to 6 hours per page |
| 3 to 12 | Corroboration: directories, associations, reviews, local press | 2 to 4 hours a month, forever |
| 6 | First honest review against the baseline grid | Half a day |
That is under 20 days of work across a year. It is not a large programme. The reason most small businesses do not get there is not the size of it, it is that months 3 to 12 have no deadline and nothing breaks if you skip them.
When AI search is not your priority
Two cases, and they are common enough to state plainly.
Referral-driven businesses with no research phase. If your customers arrive because someone told them to call you, and always have, your bottleneck is referral capture, not discovery. Fix that first.
Businesses with nothing to convert with. If the site cannot take an enquiry properly, adding visibility in front of it is spending money to fill a bucket with a hole in it. This is the same argument as search intent matching, and it applies harder at small scale where every enquiry counts.
What this costs
If you do it yourself, roughly 20 days of your own time over a year and nothing else. That is a genuine option and this page is written so it is a real one.
If you buy help, MarGen publishes its pricing rather than quoting case by case, which is worth knowing when you compare: only two of the seven UK agencies we track publish any figures at all. Read what a fair monthly price looks like before you collect quotes, and red flags when hiring before you sign one.
For proof rather than description, a small dental directory went from zero to 66,000 impressions on exactly this shape of work.
The operating side of this, meaning budget, who does the work and the monthly rhythm, is covered separately in AI search marketing for small business.
AI Search for Small Business: Common Questions
Can a small business realistically appear in AI search results?
Yes, and more easily than in traditional search, because assistants are not ranking ten results by domain authority. They are assembling one answer from sources that resolve a specific question well. A 12-person firm that answers a narrow question properly beats a large competitor that answers it vaguely. Specificity is the small business advantage and it is a real one.
Do AI models discriminate against small businesses?
Not deliberately, but the effect exists. Models lean on corroboration, and large firms have more of it: more press, more directory presence, more people writing about them. That is a corroboration gap rather than a size penalty, which matters because a corroboration gap can be closed and a size penalty cannot. Closing it takes months, not budget.
What should a small business do first?
Make your business resolvable. One canonical description, identical name and address everywhere, Companies House details matching your website, a complete Google Business Profile, and correct Organization schema. That is a day of work and it is the highest-return day you will spend. Everything else assumes the model already knows who you are.
How much content does a small business need for AI visibility?
Far less than you have been told. Ten to 15 genuinely good pages that answer the questions your buyers actually ask will outperform 60 thin ones, and 60 thin ones can actively harm you by making the model less certain what you specialise in. Publish monthly at a pace you can sustain for a year rather than quarterly in bursts.
Is AI search worth it for a business with no marketing team?
It depends on whether your buyers research before they buy. If somebody builds a shortlist before contacting you, yes. If your work arrives through referral and repeat business and always has, the honest answer is that it is a low priority and you should fix your referral capture first.
How long does it take for a small business to see results?
The foundation work can produce visible change in four to eight weeks, because most small businesses start from an entity a model cannot resolve, and fixing that removes an active blocker. Being named consistently across platforms takes three to six months. Anyone quoting faster than that on a small business is selling you the foundation work with a longer invoice.
Should a small business hire an agency or do this in-house?
Do the foundation layer yourself. It is checkable, it has a finish line, and paying a retainer for work you can complete in a week is poor value. Consider help when you hit the corroboration stage, which needs sustained outreach and a publishing cadence rather than knowledge, and which is where most small businesses stall.
What is the biggest mistake small businesses make with AI search?
Copying enterprise tactics at one twentieth of the budget. Publishing volume, broad keyword targeting and tool subscriptions are enterprise plays. At small scale they burn the budget before the foundation is done. The small business play is narrow, specific and consistent, which is less exciting and works better.
Where to Go Next
Decide, then plan: is it worth it for a small business · AI SEO for startups on a budget · build the strategy
Adjacent situations: AI search for local business · AI search for retail · does AI SEO work for brand new websites
Check where you stand: free AI visibility audit · AI Visibility Score · best GEO agency for SMEs
Have it delivered: MarGen is a UK GEO agency with published pricing, a published method and documented results.