AI visibility for small business: the cheapest way to check and fix

Start with the free AI readiness check, then fix the page that fails the plain language test. That is the cheapest way to find out where your business stands with AI search, and it costs nothing.

The rest of the job is deciding what to change and what to ignore. This guide covers the free check, the fixes that fail most often, a before and after page, and when a paid scan earns its place in the budget.

How do you check your AI search visibility for free?

Run the free readiness check before opening an AI search visibility tool. It runs fifteen checks, needs no account, and shows the fix for each one. Loosely, the fifteen checks fall into a few buckets: whether the site is reachable at all, whether the page structure is clean, and whether the business facts are actually stated. You get a list of possible obstacles without first paying to watch the obstacles sit there.

The check is a starting point, not a visibility score.

In CiteGraph's sample of 182 audited sites, the common weak points were a plain statement of what the business is, who it serves, what it costs and what it is compared against; comparison or alternatives pages; and questions answered on the page. The statement check passed on 33 sites, failed on 49 and warned on 96. The comparison check passed on 91, failed on 41 and warned on 50. The questions check passed on 99 and warned on 79. Those results do not predict your site, but they give you sensible places to look first.

None of that is a verdict on your page.

Use the free AI visibility checker after the readiness pass if you want to inspect the visibility question itself. Keep the distinction clear: readiness tells you whether a site gives an assistant usable material, while a visibility check asks what appears in answers. A clean technical site can still be absent. Fixing the technical checks does not put you in the answer.

What should you fix first on your small business website?

Start with the page a buyer would need if they asked for a recommendation. It may be your service page, product page, pricing page or a comparison page. Make the first screen answer what you sell, who it is for, what it costs or how pricing works, and what the buyer might choose instead. Put the details on the page itself, not in a button, image or script that needs extra work to read.

The fix is usually editorial before it is technical.

Use this checklist on the page you choose:

• Write a sentence that says what the business offers and who it serves.
• Put a real price, price range or clear pricing route on the page.
• Add the buyer's likely questions and answer each in a short paragraph.
• Create a comparison or alternatives page when buyers weigh you against another option.
• Give the page one clear H1 that matches its subject.
• Check that the important copy is present without JavaScript.
• Check that AI crawlers are allowed in robots.txt and that the page is reachable.

The numbers support this order. In the 182 sites checked, 178 passed the test that the homepage answers a plain request, and 175 passed the AI crawler check in robots.txt. Most sites clear those two. The harder work is making the business legible once someone reaches the page.

Reachability is rarely the problem. Legibility is.

Why do your obvious fixes fail?

The first failed fix is adding schema and stopping there. Organization schema passed on 113 of the 182 sites, while Product or SoftwareApplication schema passed on 102. Markup can confirm an entity or product, but it cannot explain why a buyer should choose the business, what the offer costs or which alternative it replaces. Another failed fix is adding an FAQ block with thin answers. Questions need answers that carry facts, limits and context, rather than a row of polite sentences that all mean “contact us”. A comparison page that compares only features has the same problem. A useful comparison explains who each option suits, where each falls short and what changes the buying decision.

It is fair to ask whether fixing your own page matters at all, if an assistant can just pull the same facts from a directory listing or a review site instead. A small business cannot control every outside page, and a clear source gives the assistant something safer to use. The answer is not to polish one page forever. It is to remove the avoidable gaps, then measure whether the result appears in the questions that matter.

This is where the work gets expensive without getting useful. More pages do not repair a missing statement. More schema does not create a comparison. More FAQ headings do not make an answer specific.

How should you rewrite a page so AI can use it?

Imagine a service page that currently says the business is “helping organisations improve through tailored solutions”. Replace it with a direct statement such as: “We provide bookkeeping for small UK retailers, with monthly packages from $149, including up to 200 written pages.” Follow it with the types of retailer served, what the monthly work includes, what is not included, how the price changes, and the alternatives a buyer might consider. Add questions about timing, setup, support and cancellation, with answers that stand on their own. This is a hypothetical example, not a reported client result.

Before, the page describes an ambition. After, it supplies a buying answer.

Keep the concrete facts in normal page text. The title and description check passed on 137 of 182 sites, and the one clear H1 check passed on 151. Those are useful foundations, but neither tells an assistant everything it needs to know. Canonical and Open Graph tags passed on 140, while 38 warned. These tags help a page present itself correctly, but they do not replace a clear offer.

Use the page to answer the request a buyer is likely to make, not the request you wish they had made. “Who is this for?” is more useful than “Why are we different?” when the page has not yet established what it sells. “What does it cost?” is more useful than a slogan when the price is available elsewhere but not here.

How do you test whether ChatGPT recommends you, without fooling yourself?

Write a small set of real buyer questions. Include a category question, a fit question, a price question, an alternatives question and a question that names a competitor or substitute. Run them more than once if you can, because one answer is an observation, not a verdict. Record whether your business is named, whether your site is cited, which page is cited and which competing pages appear instead.

Track the citation, not just the mention.

A business can be named without its own page being used. In the 182 site sample, the check for the site's own pages getting cited passed on 25 sites, failed on 19 and warned on 7. That is a small pass count beside the broader technical checks. It is a reminder to inspect the source page and the answer separately.

You can also use the live check of which AI engine a site is built for when the technical question is the priority. For a fuller explanation of the measurement, see how CiteGraph measures AI visibility. We measure samples; we print margins when the study has them; we do not know what we have not measured. A handful of manual prompts cannot establish a market position.

When should you pay for an AI search visibility checker?

Pay for a scan when you have a defined set of buyer questions and a reason to watch them more than once. That reason might be a launch, a new pricing page, a competitor change or a decision about which pages to improve next. If you have not fixed the obvious statement, pricing and comparison gaps, a paid tracker may give you a tidy report about a messy page. Tidy reports are pleasant. They are not a substitute for the page.

Start small if the work is small.

CiteGraph's pricing page lists Starter at $29 a month for 2 projects, 10 questions each, 3 runs per question on 4 engines, 6 tracked rivals and 30 written pages a month. Growth is $69 a month for 3 projects, 10 questions each, 5 runs per question on 4 engines, 10 tracked rivals and 60 written pages a month. Scale is $149 a month for 6 projects, 10 questions each, 5 runs per question on 4 engines, 10 tracked rivals and 200 written pages a month. Every plan starts with a 7-day trial opened with a $1 card confirmation and includes 2 scans. A single scan without a subscription is $12, according to the CiteGraph pricing page.

For an owner with one site and one afternoon, a single scan may be enough to compare the starting state with the page after the fixes. A subscription makes more sense when you will repeat the same questions, compare rivals or report changes to someone else. The useful unit is not the dashboard. It is the decision the scan helps you make.

What should you do this afternoon?

Set aside the afternoon in four blocks. First, run the free check and save the results. Second, choose a buyer-facing page and repair the plain statement, price route, questions and comparison context. Third, run your real buyer questions and record names, citations and source pages. Fourth, decide whether a one-off scan or a recurring tracker will change the next decision.

Do the page before the purchase.

If your team works in a repository, CiteGraph can run as an MCP server so Claude, Cursor or VS Code can read a scan and apply readiness fixes in a repo. That may be useful when the bottleneck is implementation. If the bottleneck is knowing what to write, start with the page and the buyer question instead.

The cheapest check is still the free one. What changes after the afternoon is not that your business has earned a guaranteed place in an AI answer. You have removed specific obstacles, tested the questions that matter and learned whether paid measurement has a job to do.