What GEO really costs, and what we refuse to sell

by | Aug 26, 2026 | GEO

You ask for a quote. You get back a range, a monthly retainer, a six or twelve month commitment. You get told about proprietary methodology, continuous monitoring, iterative optimization.

What you almost never get told: how much, for what exactly, and how you will know it worked.

The problem is not that vendors are dishonest. The problem is that a market with no measurable deliverable has no reference price, and a market with no reference price charges whatever the client agrees to pay.

This article gives you the only serious public figures on what companies actually spend, our own rates in full, and the list of things we refuse to bill for.

What the market spends

The only public survey with a declared sample dates from March 25, 2026. It covers 205 executives and marketing leads at US organizations, supplemented by interviews with several agencies.

Provenance marking: commissioned survey, relayed by the trade press, sample size announced, methodology thin. That is weak as evidence, and it is still the best that exists.

Fifty-five percent of respondents now have a dedicated budget line. Among them, seven in ten allocate between eleven and twenty percent of their budget to it. One agency quoted in the survey recommends a pilot budget worth one and a half to two times the existing paid search budget.

Two details in that same survey are worth more than any of the percentages.

Not one of the practitioners interviewed disclosed an amount. Not in dollars, not in euros. In a sector where campaign rates, copywriting rates and audit rates have circulated freely for twenty years, this one publishes nothing.

And one of the executives quoted describes the real mechanism: if you were spending ten thousand dollars a month on search, half of that now covers GEO as well.

Translation: in a good share of cases, this is not a new budget for new work. It is the same budget, for the same work, under a new name.

On failure rates, there is nothing to show you

We went looking for data on the proportion of GEO engagements that miss their stated objective. There is no serious study.

What circulates instead is self-promotional material published by the vendors themselves: three hundred and forty percent return on investment, three hundred and sixty-six percent visibility gained. Provenance marking on all of it: vendor self-reported, no methodology, no sample, no control group.

Those figures will not be quoted here, not even to attack them. Giving a number like that something to be remembered by is already giving the vendor what they came for.

The absence of any data on failure, in a market reporting 55 percent adoption, is information in itself.

Our rates, in plain text

We have published our prices since the beginning, and this subject gets no exception.

Service	Price	What it includes
Technical audit	200 dollars	Detailed report, prioritized by revenue impact, plus a live walkthrough over video
Content writing	200 dollars for 20 articles	No commitment, renewed monthly
Product listing	5 dollars each	Whatever the platform
Community management	100 dollars per month	Three posts per week, with photo or video
Advertising	200 dollars setup, then 100 per month	All platforms, you pay the ad network directly
Website	From 1,000 dollars	Up to ten pages, six months of unlimited support
Hosting	From 149 euros per year	4 GB dedicated, daily backups, 99.9 percent SLA

You will notice there is no line called GEO in that table.

That is not an oversight. The actions the literature reports an effect for, meaning real coverage of a topic, presence on third-party sources, and correction of the information circulating about you, are already covered by content writing, community management and the audit. Adding a separate line would mean billing the same work twice, which is exactly what the survey above describes happening across the market.

What we refuse to sell, and why

A guaranteed position in ChatGPT or in a Google summary. There is no stable ranking to guarantee. The same question asked twice does not produce the same answer: generative visibility is a distribution, not a point. Anyone guaranteeing a position is selling a thing that does not exist.

A setup package for an llms.txt file. The file takes ten minutes to deploy. An analysis of roughly 300,000 domains finds no correlation between its presence and citations, and the model tested actually performs better without it. Another, covering 137,000 sites, measures that 97 percent of those files are never read at all. Provenance marking on both: SEO tool vendors reporting on their own market. We will deploy the file free of charge if it matters to you. We will not invoice it.

Schema markup sold as a citation lever. Markup has genuine value for crawling, interpretation and rich results. But a study covering 1,885 pages that added markup, compared against roughly 4,000 control pages, measures no effect at all on citations from two engines, and a statistically significant decline on the third. Provenance marking: SEO tool vendor, with a control group, which is more rigor than most of this field offers. We bill markup for what it does, not for what people wish it did.

A monthly report built on a single measurement. A number obtained by asking an engine once is a draw, not a measurement. We will not produce a chart that implies otherwise.

A visibility percentage promised in advance. The gain from a generic strategy collapses from 0.802 to 0.007 as soon as competitors adopt the same approach, according to an arXiv preprint modeling competitive adoption. Promising a percentage means betting that your sector will stand still.

Rewriting “for the LLMs” with no measurement of retrieval. The critical survey that examined forty-five studies, published as an arXiv preprint in July 2026, reports that these rewrites can degrade retrieval of the document itself. Optimizing one stage while breaking the previous one is not a service. It is a transfer of risk onto you.

The three-question test

Faced with any quote at all, ask these three questions. None of them requires technical knowledge.

What happens if I do not buy this? If the answer is “you will be invisible in AI”, ask for the measurement of your current visibility. You are probably not at zero, and the burden of showing otherwise sits with the person asking for money.

How many times do you measure, and against which control group? A single measurement, with no control, supports no attribution whatsoever. You will be paying for platform variance dressed up as progress.

What share of the work happens off my own site? If the answer is zero, the engagement is aimed at the place where the measurements report the weakest effect.

A serious vendor answers all three. A salesperson changes the subject on the second.

The calculation nobody offers you

Before you sign, run this calculation. It takes fifteen minutes.

Pull the share of your sessions arriving from generative platforms. The average measured across 1,215 domains is 1.08 percent, per a vendor benchmark report, and yours may be very different depending on your sector.

Apply your conversion rate and your average order value to it. You now have the revenue currently attributable to that channel.

Add a conservative estimate of the indirect effect, the one that travels through a later brand search and that your analytics tool files under organic. A June 2026 panel study, published as an arXiv preprint, measures a 4.3 point rise in brand search after a recommendation.

Compare that against the annual cost of the engagement.

If nobody, neither you nor the vendor, can run that calculation with defensible orders of magnitude, then you are not considering an investment. You are considering a bet, and you should at least know that you are placing one.

What you do tomorrow morning

Pull up the last GEO quote you received and strike out every line that appears in our refusal list above. Look at what survives, and at the price attached to it.

Then ask what that remaining price buys for the only thing that matters: being present, accurately, in the sources the engines actually cite in your category.

There is no GEO line on our invoices. There is writing, presence, measurement, and thirty years of habit of not billing for what we cannot demonstrate.

Sources

<strong>LaFactory</strong> measures AI visibility with a published protocol: repeated measurements, paraphrases, control group. No guaranteed placement, ever. Contact us to scope an audit.