Short answer, as of September 2026: with two hours a month you can hold a real position in AI answers — spend roughly six hours once on eight foundational pages, then thirty minutes a month checking a dozen prompts and ninety minutes fixing whatever the check surfaces. What you cannot do at that budget is compete on volume, and for most small businesses volume was never the lever anyway.
Nearly all GEO advice is written for teams with a content function. The owner-operator reading it concludes the whole thing is out of reach and does nothing, which is the worst available outcome — because the things that actually get a small business cited are mostly one-off fixes rather than ongoing production.
Here is the minimum that works.
The one-off: eight pages, about six hours
These are not blog posts. They are the pages an assistant needs in order to describe you accurately, and once they are right they mostly stay right.
| Page | What it must state plainly | Why an assistant needs it |
|---|---|---|
| What you do | Services in the words customers use, not internal names | Category placement |
| Who you serve | Segments, and who you are not for | Qualification answers |
| Where you work | Service area or locations, explicitly | Local and "near me" style answers |
| What it costs | Real ranges, and what moves the price | Cost prompts, the most common kind |
| How it works | The process from enquiry to completion | "What happens if I" answers |
| What you do not do | Honest limits and exclusions | Fit and comparison answers |
| Who you are | Credentials, years, licences, ownership | Trust and entity resolution |
| How to reach you | Consistent name, address, phone, hours | Entity consistency everywhere |
Two rules about how these are written matter more than the list itself.
Write in self-contained sentences. An assistant quotes passages, not pages. "We serve the metro area and surrounding counties" is useless out of context; "Acme Plumbing serves the Springfield metro area and the three surrounding counties, with no call-out charge inside a twenty-mile radius" survives being lifted. The habit of making every important sentence stand alone is the single highest-return writing change available.
Put the answer before the story. The paragraph that answers the question should come first, with background after. Pages that build to the answer get skipped by retrieval and by readers.
If your site is new and has no authority yet, the sequencing is slightly different and worth reading: GEO for a brand-new site with no authority.
The thing most small businesses get wrong
Pricing. It is the most common question asked of assistants about any service business, and the most commonly withheld answer.
Withholding does not prevent the question being answered. It means the answer gets assembled from competitors, directories and forum guesses — usually inaccurately and rarely in your favour. Publishing a real range with the variables named ("typically $X to $Y depending on A and B") makes you the source of your own pricing. The full argument, including what to publish when every job is custom, is in publishing your prices.
This is the highest-leverage hour in the entire six.
The monthly check: thirty minutes
Pick twelve prompts — the questions your customers actually ask, phrased the way they would type them. Not keywords. Sentences.
A workable twelve for most service businesses: what you do plus your area; the cost question; "best [service] in [area]"; two specific problem questions you get asked weekly; "who should I call for [problem]"; a comparison against the obvious alternative; an urgency version ("emergency [service] near me"); a qualification question ("do I need [service] or [other service]"); your own business name; a competitor's name; and one question about a thing you do that competitors do not.
Ask those twelve in two assistants. Record three things per prompt: were you named, what were you called, which sources were cited.
That is the whole check, and it takes about half an hour once you have the list written down. Keep it in one file with a row per month. The value is entirely in the trend — a single month's snapshot tells you almost nothing, because answers vary between sessions.
At twelve prompts and two engines, doing this by hand is perfectly reasonable. It stops being reasonable at around twenty prompts or four engines, which is the point at which software earns its cost rather than before; the categories and their price points are laid out in AI visibility tools: the five categories.
The monthly fix: ninety minutes
The check produces one of four findings. Each has an obvious response.
You are not named at all. Before writing anything, verify that your content is readable — that assistants can reach the site and that the key facts are in the HTML rather than assembled in the browser. A surprising share of "we are invisible" cases are access problems, not content problems, and it is a fifteen-minute check: AI crawler access: the robots.txt and WAF audit.
You are named but described wrongly. Find the source of the wrong fact — usually an old directory listing, a stale profile, or a page on your own site you forgot about — and correct it there. Correcting it only on your homepage rarely works if the error lives somewhere the engine trusts more.
Someone else is named for a question you should own. Read what the cited source actually says. Usually it answers the question more directly than your page does. Rewrite your page to answer it in the first paragraph.
A prompt went from named to not named. Check whether the page that earned it changed, or whether a competitor published something better. Citations decay for predictable reasons, and the diagnostic sequence is in citation decay: why a page that was cited stops being cited.
Ninety minutes is enough for one of these per month. Do the highest-value prompt, not the easiest.
What to skip
Being deliberate about what you are not doing is what makes two hours sustainable.
Skip publishing on a schedule. A monthly blog post written to fill a slot is the least efficient GEO activity available to a small business. One excellent page that answers a real question beats twelve thin ones, and it keeps earning without maintenance.
Skip chasing every engine. Two is plenty. Pick the ones your customers use and ignore the rest until something changes.
Skip social media as a GEO tactic. It is a fine channel for other reasons; it is a weak citation source.
Skip tracking rankings alongside this. Different system, different work. Mixing them turns a thirty-minute check into an afternoon.
Skip rewriting pages that are working. The instinct to "refresh" a page that is being cited is how small businesses accidentally break their best asset.
A realistic first ninety days
Month one: write or fix the eight pages. This is the six hours and it is front-loaded on purpose — the point of the routine is that the heavy part happens once.
Month two: build the twelve-prompt list, run the first check, and do nothing else. You are establishing a baseline. Resist acting on a single reading, because you cannot yet tell a real gap from normal variance.
Month three: run the check again, compare, and fix one thing. This is the first month the routine is actually running, and from here it stays at two hours.
By month six you will have four data points and a clear sense of which two or three prompts you are genuinely competitive on. That is a defensible position, and it was bought with about twelve hours of work in total.
When two hours is genuinely not enough
Being straight about the ceiling: this routine holds a position in a market where you have a real business and a normal number of competitors. It does not work in three situations.
If you operate across many locations, the prompt set multiplies immediately and per-location visibility diverges sharply from the national picture — the mechanics are in tracking AI visibility across every location.
If you have no physical premises and compete on service area, the entity work is harder than average and worth reading up on: AI visibility for service-area businesses with no storefront.
And if you are trying to displace an established competitor rather than hold your own ground, that is a content programme, not a maintenance routine. Two hours a month will not do it, and no honest tool or agency will tell you otherwise.
Frequently asked questions
Can a small business do GEO without hiring anyone? Yes. The work that gets a small business cited is mostly one-off — eight pages that state plainly what you do, who for, where, at what cost, and what you do not do. After that a monthly check of a dozen prompts and one fix keeps it current. Roughly six hours once, then two hours a month.
How many prompts should a small business track? About twelve, phrased as full questions rather than keywords, across two assistants. That is enough to detect real change and small enough to check by hand in half an hour. Software becomes worth its cost at around twenty prompts or four engines.
What is the single highest-impact thing to publish? A real pricing page with ranges and the variables that move them. Cost is the most common question asked about service businesses, and it gets answered whether or not you participate — withholding it means the answer is assembled from competitors and directories instead of from you.
Do I need to publish blog posts regularly for AI search? No. A monthly post written to fill a slot is the least efficient option available at this budget. One page that answers a real question thoroughly outperforms twelve thin ones and keeps earning citations without maintenance.
How long before a small business sees results from GEO? Expect the first movement on narrow, specific prompts within six to twelve weeks of fixing the foundational pages. Broad category prompts take much longer and may never be winnable against larger competitors, which is why the routine concentrates on specific questions.
What should I check first if AI never mentions my business? Whether assistants can actually read your site — crawler access, and whether your key facts are present in the HTML rather than assembled by JavaScript in the browser. A meaningful share of invisibility cases are access problems rather than content problems, and it is a fifteen-minute check.
Is two hours a month enough if I have several locations? No. Each location needs its own prompts, because national visibility routinely hides markets where you are never named. Multi-location operators should budget for per-location tracking from the start rather than scaling this routine up by hand.