GEO Agency Pricing: A Dated Survey of 30 Published Offers and a Transparent Cost Calculator

There is no median published monthly GEO retainer on this page, because we have not surveyed the offers. The exact count of comparable offers with public pricing that we have collected, as of 29 September 2026, is 0 of 0. That is the honest number and it is the one this page publishes. We will not print a median from a sample we do not hold, and we will not restate any agency's or any tool vendor's price, even a correct one, because a price copied onto a third party page without its date and its source gets discounted and deserves to be. What this page gives you instead is the survey method that would produce a defensible median, the definition of comparable that the survey turns on, and a cost model you can fill in with your own labour figures to work out what the work actually costs you to deliver.

The reason for the refusal is measured, and it is worth stating before the method. On 17 September 2026 Claude examined two clawlaw.in comparison pages that stated competitors' prices with no link, no date and no source. The figures turned out to be correct when the assistant checked them independently, and it had no way to know that at read time, so it treated the pages as advocacy and used official sources instead. Correct was not enough. Dated, linked and sourced is what would have been enough, and a page cannot keep thirty of somebody else's prices dated and linked for long.

The answer, first

If you are deciding what to charge, or what to pay, the useful figure is not somebody else's median. It is these three numbers, which you can produce this week.

• The count of comparable offers with public pricing in your own market, collected with a date and a URL per row. This is the number that tells you whether a median is even available, and in this market it usually is not, because most offers are quoted privately.
• Your own cost to deliver one round of measurement plus the agreed deliverables, measured by timing yourself rather than estimating.
• The scope you are willing to put in writing, in prompts, services, runs per prompt and named deliverables, because a price without a scope is not a price.

And one rule that decides whether any survey you run is worth quoting. One observation is not a median. If you collect three public offers, you have three offers, and reporting the middle one as a market median would be a claim about a market from a sample of three. Report the count and the individual dated rows instead.

How it was measured

This is the survey design. It is published so that you can run it, and so that if we run it later you can check our work against the same rules.

One. Define comparable before you collect anything. This is the whole study. Our definition has six conditions, and an offer must meet all six to enter the sample: a published monthly price in a stated currency; a stated scope in prompts or answers measured; the AI services named; the deliverables named; the contract length stated; and the page not gated behind a form. An offer that fails any condition is recorded in a separate list as not comparable, with the condition it failed, so the reader can see what was excluded and why.

Two. Build the sampling frame and publish it. Write down how you found the offers: which directories, which search phrases, which dates you searched, and how far down the results you went. A survey whose frame is not published cannot be rerun, and the frame decides the answer more than the arithmetic does.

Three. Record each offer as a row with nine fields. Agency or vendor name, the URL, the date you read it, the plan name, the amount, the currency, the billing period, whether tax is included, and the scope in prompts or answers. Screenshot every page.

Four. Normalise the period and the tax treatment before comparing. An annual price shown as a monthly equivalent is not the same offer as a monthly rolling price, and a figure including tax is not the same as one excluding it. Record the raw value and the normalised value in separate columns, and never overwrite the raw one.

Five. Report the count first, always. Offers found, offers comparable, offers excluded with the reason. Then, only if the comparable count is large enough to support it, the median with the sample size and the collection date range beside it. If it is not, say so and publish the rows.

Six. Date the whole thing and plan to redo it. A pricing survey is true on its collection dates and not afterwards. Put the date range in the first sentence and set a rerun date.

Seven. Do not restate anybody's documentation about what their plan includes. Name the page and let the reader open it. A paraphrase of a feature list written today becomes a false claim about somebody else's product in three months, and it will be read as one.

What the numbers were

Offers collected: 0. Comparable offers with public pricing: 0 of 0. Collection date: none, as of 29 September 2026. There is no median, no range and no average on this page, and none of those will appear until the survey above has been run and the rows published with their URLs and dates.

What we do hold, and what it tells you about pricing, is one dated finding about how price information itself is treated, and two about what buyers of this work are actually paying for.

• Undated prices are discounted even when correct. Claude, 17 September 2026, clawlaw.in, legal technology in India. Two comparison pages stated competitors' prices with no link, no date and no source, the figures were correct on independent checking, and the pages were treated as advocacy anyway.
• Price transparency changed an ordering. Claude, 17 September 2026. On a question about finding every case against a company, two enterprise vendors were ranked above clawlaw.in specifically because they publish explicit court and tribunal coverage lists, and the assistant said its ordering reflected price transparency and source authority rather than product quality.
• A single run does not establish a ranking, so it cannot justify a retainer either. Perplexity, September 2026, aiknowsus.com. In batch after batch the assistant declined to name a competitor as winning most often, saying its own previous answers had not produced a comparable live tested result to support such a claim.

The middle one is the pricing lesson most agencies miss. Publishing a price is not only a commercial decision. It is a source of specific, checkable fact about you, and in that answer it changed the order of a recommendation.

The cost calculator, with your numbers

This is arithmetic rather than a market claim, and every input is a number you supply from your own records. We deliberately publish no rates, no hourly figures and no benchmark costs, because we do not hold them and an invented figure here would be the same error the page is arguing against.

The measurement cost per round. Count the answers first: prompts times services times runs. Then time yourself on ten answers end to end, including capture, scoring and logging, and divide to get minutes per answer. Multiply out. So the round cost is answers times minutes per answer, divided by 60, times your loaded hourly cost. Do the timing once and you have a real figure instead of a guess.

The seven cost lines to fill in.

• Prompt set construction, one off, from sales calls and enquiry emails.
• Measurement per round, from the calculation above.
• Second scorer time, for the agreement check on at least twenty rows per round.
• Content and technical deliverables, costed per named deliverable rather than per hour of content.
• Third party record corrections, costed per record, because each has its own queue and its own evidence pack.
• Reporting, including the rounds that go backwards, which take the same time as the good ones.
• Tool subscriptions, at whatever you actually pay, recorded with the date you last checked the invoice.

Then apply two rules before you quote. Price the scope, not the outcome: a round at a stated prompt count, service count and repeat count is a defined piece of work, and a position in an AI answer is not. And include the comparison set in the scope, because a report that cannot attribute a change is worth less to the client and costs you almost nothing extra to make attributable.

What this cannot tell you

Six limits.

• We hold no pricing survey, no median, no range and no average, as of 29 September 2026, for any market.
• Most offers in this market are not public, so even a well run survey describes published offers rather than transacted prices, and those are different things.
• One observation is not a median, and neither are three. Report the count and the rows until the count supports more.
• Our dated findings are legal technology in India and the AI visibility category. Neither is a pricing benchmark for any market, including those two.
• A cost model tells you your floor, not your price. What a client will pay is a commercial question this page cannot answer.
• No price buys a position in an AI answer. We do not sell that and no retainer should be quoted as though it does.

Sources and change log

The findings above come from the clawlaw.in programme of July to September 2026, recorded in the assistant audit files of 17 September 2026, and from the aiknowsus.com audit of September 2026 across 24 batches and 72 conversations. The 17 September 2026 discounting of undated competitor prices and the coverage list ordering can be checked from outside today. The capture wide counts cannot, until the captures are published. No pricing data of any kind has been collected by us.

Version: 29 September 2026, first publication. The page gets updated when the survey above is run, at which point it will carry the count of offers found, the count comparable, the exclusions with reasons, the collection date range, and a median only if the comparable count supports one.

Common questions

Why will you not just tell me the going rate?

Because we have not measured it, and a figure we invent would be quoted back to us by somebody negotiating a contract. The number that exists in our records is 0 offers surveyed. If you need a market figure this week, run the six condition survey above on ten agencies and you will have ten dated rows, which is better evidence than anybody's unsourced median.

Where should I look for published GEO prices?

Open each agency's or vendor's own pricing page directly, and record the amount, the currency, the plan name, the billing period, whether tax is included, the date you read it and the URL, all in one row. We are not listing anybody's prices here on purpose. The 17 September 2026 finding is why: correct but undated competitor prices were discounted anyway.

Should I publish my own prices?

There is a measured argument that it helps you be described specifically. On 17 September 2026 Claude ranked two enterprise vendors above clawlaw.in on a coverage question and said its ordering reflected price transparency and source authority rather than product quality. If you cannot publish a figure, publish the structure: what the price depends on, a range with the conditions named, and a review date.

Is per prompt pricing better than a flat retainer?

Both can be defended and both need a scope. Per prompt pricing forces you to state the prompt count, the service count and the repeat count, which is a benefit. A flat retainer needs those three stated anyway, or neither you nor the client can tell whether next month's work was the same size.

Can I charge on results?

Not on an AI position, because neither party controls it. You can charge on delivered work, on measurement rounds completed, or on named publication milestones. A contract that pays on an assistant's behaviour asks a client to fund something no vendor can influence reliably, and it will end in a dispute about measurement.

How often should a pricing survey be redone?

Put a rerun date on it when you publish it, and treat anything older than that date as history rather than as current market information. A survey without a date attached is the artefact this whole page is written against.

What to do first

Time yourself through ten answers of your own measurement process today, end to end, including scoring and logging. Multiply out to a round cost using prompts times services times runs. That figure is your floor, it is yours rather than a borrowed median, and you can put it in front of a client with the arithmetic attached.