What GEO should cost a single-location cafe, and when you should not buy it

Cafe owners tend to ask this question with a quote already in hand. Someone has sent a monthly figure with the words "AI visibility" attached, and they want to know if it is fair. We are not going to put a number on it here, because any number would be wrong for half the people reading. What we can do is describe the shape of the spend, and the shape is clear. For a single-location cafe, most of the value in GEO sits in a one-off technical fix and a round of listings hygiene. A retainer is the exception, and it needs a specific justification to exist.

We sell GEO. That is a reason to read the rest of this with some scepticism, and it is also why we would rather say it plainly than have you find out after a year of invoices.

What the published rates say, and who wrote them

Here are the numbers, because a page about cost that hides the cost is doing the thing it is complaining about.

Now the part that matters more than the figures.

Every one of those ranges was published by a company selling GEO. We fetched RevvGrowth's pricing page, which quotes $3,000 to $25,000 a month and sells a $3,000 Starter and an $8,000 Growth plan on the same page. It cites no source and no methodology for the range.

That is not an accusation, it is the structure of the whole category. There is no independent survey of GEO pricing. There is a set of agencies each publishing the band its own packages sit inside.

We are one of them, which is why the useful thing we can give you is not a number but a test: ask what is being done, on which week, and how you would know it happened. A retainer that cannot answer that is a subscription to a report.

And a cafe is not any of the tiers above. The honest scope for one location is a project, and the rest of this page is what goes in it.

What GEO should cost a single-location cafe: the shape of it

The work splits into pieces that finish and pieces that do not. Almost everything with a measurable effect belongs in the first group.

Read down the middle column. One row continues. That row is the only honest case for a retainer, and we will come back to it.

Why the technical fix is a project and not a subscription

Each item in the finishing group has a defined end state. Either the homepage reads correctly with JavaScript off or it does not. Either the old menu URL redirects cleanly or it returns a 404.

Once these are done they stay done until someone redesigns the site or swaps the platform. Paying monthly for them is paying rent on a finished job. If a proposal lists "technical GEO" as a recurring line, ask what changes month to month. The honest answer is usually nothing.

There is one piece of this a cafe owner can check alone. Open your homepage, disable JavaScript in the browser, reload. If the menu and the hours disappear, ChatGPT, Claude and Perplexity see an empty page. Only Google's AI surfaces and Microsoft Copilot render JavaScript. Fixing that is the first invoice, and for many cafes it should be the only one for a while. We cover the crawler side in more detail in our guide to AI crawlers and robots.txt.

Listings before anything with "AI" in the name

Local SEO comes first, and for a cafe it is most of the job.

Google's AI Overviews and AI Mode run on Google's own index. For a cafe, the Business Profile is the local record Google holds on you, so it needs to be complete and current. We wrote about that separately in AI Overviews for local businesses. Copilot grounds on Bing, so Bing Places matters more than its traffic suggests.

ChatGPT is the interesting one. An independent test of 2,880 prompts in August 2026 found Yelp grounding in about 96% of ChatGPT local runs, after OpenAI's non-exclusive licence deal with Yelp in July 2026. Foursquare appeared in 0% of them, which retires the story that Foursquare powers ChatGPT local results.

For a cafe that means the Yelp page is doing more work than the website in a good share of ChatGPT answers. Wrong hours on Yelp, or a category set to "Restaurant" when you are a coffee shop, gets read straight into the answer. Correcting it needs a login, not an agency.

When a retainer earns its keep

The continuing row in the table is third-party coverage, and here the evidence is consistent even if it is only correlational.

Ahrefs looked at 75,000 brands and found branded web mentions correlated with AI Overview visibility at 0.664, against 0.218 for backlinks. Muck Rack analysed more than 25 million citations and found earned media accounted for 84% of them, with paid and advertorial placements at 0.3%. Neither study proves cause. Both point the same way.

For a cafe, earned coverage means the city guide, the neighbourhood newsletter, the food writer's "where to get a flat white near the station" list, the local paper's weekend round-up. These get written continuously and they go stale continuously. That is work with no end state, and that is what a retainer is for.

The test for whether the retainer is real: can the person show you the placements each month, by name, with a link. Pitching a cafe to a writer and following up until it lands in a list is labour. "Monitoring" is not. If the monthly deliverable is a dashboard, you are paying for a screenshot.

One caution. Being cited and being named are different things. Semrush's ghost citations study found 61.7% of AI citations produced no brand mention at all, and ChatGPT cited a source 87% of the time while naming a brand 20.7% of the time. The coverage you want names the cafe in the body of the piece, not in a link at the bottom.

When you should not buy GEO at all

• Your Google Business Profile is incomplete or your reviews sit unanswered. Do that first. It is free and it is the local record Google holds on you.
• Your site is a builder page that draws its content in the browser and you are not willing to change platform. GEO spend on top of that is wasted until the rendering changes.
• The proposal's deliverables are llms.txt, schema markup or FAQ markup. Google states it does not use llms.txt. Ahrefs' matched study of 1,885 pages that added JSON-LD found AI Overview citations fell 4.6%, with ChatGPT and AI Mode changes indistinguishable from zero. FAQ rich results ended in Google Search on 7 May 2026. Our schema for AI search piece has the full argument.
• The pitch includes a "visibility score" built from a few prompt runs. LLM outputs are non-deterministic. Five repetitions of a prompt carries a margin of error around plus or minus 27 points. Sampling needs 12 or more runs per prompt before the number means anything.
• You expect it to show up in web traffic. AI referrals sit around 1% of traffic in Conductor's benchmark, and many arrive with no referrer and land as Direct. For a cafe the outcome is a person at the counter, which no analytics tool attributes.
• Anyone guarantees placement. There is no submission form and no index status page, so a guarantee is a guess with a price on it.

How to check you got what you paid for

Bing Webmaster Tools has an AI Performance report, free since February 2026, which shows Copilot citations and the grounding queries the model generated. Google Search Console's Generative AI report shows impressions only. Beyond those, the honest method is a fixed set of prompts a real customer would type, run at least 12 times each, sampled over time. We lay that out in how to track AI search visibility.

If the first report does not include a before-and-after of the JavaScript-off homepage test and a list of listing corrections made, the technical work did not happen.

Before you sign anything, open your Yelp page and check the hours against the door. If they are wrong, that is most likely what ChatGPT is reading.