Marketing agency pricing in 2026: a dated, auditable survey of published fees and project budgets

Almost no agency publishes a price, which is why every pricing article in this field quotes other pricing articles. The honest number to lead with here is the count of dated public agency price pages reviewed, and as of 29 September 2026 ours is zero. So this page publishes no fee figures at all. What it does publish is the whole survey: how to build the sampling frame, what counts as a published price, the thirteen fields to capture per page, the six pricing models to code against, the scope of market and service, and the exact form the results will take. Any reader can run it, and when we run it the count and the date will appear in the change log below, next to this sentence.

The answer, first

Three statements, each one checkable against the rest of this page.

We have reviewed zero dated public price pages, so we publish no prices. Not a range, not a median, not a "typical retainer". A price range with no sample behind it is the most copied invented number in this sector, and the moment we printed one this page would be worth less than the silence.

The market and service scope is fixed in advance. India, agencies selling to Indian businesses, and six named services: search optimisation, paid media management, social media management, content and editorial, web design and build, and AI search or generative engine optimisation. Anything outside that is excluded and counted as excluded.

Unsourced prices are not a shortcut, they are a penalty. On 17 September 2026, in our work on clawlaw.in, two comparison pages stated competitors' prices with no link, no date and no source. Claude checked the figures independently and they turned out to be correct, and it still treated the pages as advocacy and used official sources instead, because it had no way to know they were correct at read time. That is the entire argument for doing this the slow way.

How it was measured

Step one: build the sampling frame before looking at any price. Write the list of agencies first, from sources you can name, so nobody can accuse you of picking the prices that suited your story. Four sources, combined and de duplicated:

• A named industry association membership list, such as the Advertising Agencies Association of India.
• Two named agency directories, recording the filter you used, for example the country and service filters on a directory such as Clutch.
• The agencies named in AI assistants' own answers to a fixed set of ten buyer queries, recorded with the date and the assistant.
• The first five pages of results for four fixed search queries, recorded with the date.

Publish the frame size and how many came from each source, then the de duplicated total. That total is your denominator for coverage, and it is a different denominator from the number of agencies that publish a price.

Step two: define a published price so the decision is not a judgement call. A page qualifies only if it carries all four of these: a number or a band, a currency, a unit the number applies to, and a page you can archive. Excluded, and counted separately: "starting from" with no unit, "contact us", a price shown only after a form, a price in a downloadable PDF behind an email gate, and a price on a third party directory profile rather than the agency's own site.

Step three: capture thirteen fields per qualifying page. This is what makes the survey auditable rather than merely dated.

• Agency name, and the source that put it in the frame.
• The exact URL of the price page.
• Date captured, in ISO form.
• An archive link to the captured page.
• Whether the page itself carries a date, and what it is.
• Service, coded to one of the six.
• Pricing model, coded to one of the six below.
• The figure or band, exactly as printed.
• Currency.
• The unit: per month, per project, per hour, per deliverable.
• Whether tax treatment is stated.
• Minimum term or minimum commitment, if stated.
• What the price includes and excludes, quoted from the page.

Step four: code the pricing model to one of six. Monthly retainer. Fixed project fee. Hourly or day rate. Percentage of media spend. Performance or outcome based. Hybrid, meaning a retainer plus one of the others, with the components recorded separately.

Step five: two people capture the same fifty pages independently, and the number of disagreements gets published. Most disagreements in a pricing survey are about whether something is a price at all, which is why step two is written as four hard conditions.

Step six: report counts before anything else. The frame size, the number of pages checked, the number that qualified, and the number that did not with the reason. Only then a distribution, and only as a median with the sample size beside it, never as an average of bands. Never a single "typical" figure.

Step seven: date stamp the whole thing and set a recheck date. Prices change without an announcement, and an undated price survey is a wrong price survey within a year.

What the numbers were

Dated public agency price pages reviewed: zero, as of 29 September 2026. The frame has not been built, no pages have been captured, and no figure is published on this page. That is the number the page leads with, and it will be replaced by a real one in the change log when the survey runs.

What we deliberately refuse to do instead. We will not restate figures from other pricing articles. We will not average bands we have not seen. We will not convert somebody's dollar range into rupees and present it as an India figure. Each of those would produce a number that looks exactly like a measurement, and the reason to refuse is in the record: on 17 September 2026 correct but unsourced competitor prices on two clawlaw.in pages were still discounted as advocacy by Claude, which used official sources instead.

The dated results we do hold about prices and how assistants treat them.

• On 6 August 2026 the clawlaw.in pricing page rendered its prices only after scripts had run, so a crawler received no prices at all, and the prices the assistants quoted had come from an app store listing rather than from the company's own site. Claude's run recorded it.
• The same day, ChatGPT found that the website and the app store listing carried different plan names and different prices for the same product, and it said so inside its answer.
• 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.

Read together, those three are the practical finding for any agency reading this page. Price transparency is treated as a source quality signal, a price that only a script can render is a price that does not exist, and two public price lists that disagree will be quoted against you.

What this cannot tell you

• It cannot tell you what agencies actually charge. A survey of published prices is a survey of the agencies willing to publish, which is a small and unusual group.
• It cannot tell you what you will be quoted. Quotations depend on scope, seniority, market and how badly the agency wants the work.
• It cannot compare a retainer with a project fee without a defined scope, and scope definitions differ between agencies by more than the prices do.
• It cannot stay true. Every figure it would produce needs a capture date, and the survey needs rerunning rather than updating.
• It cannot tell you value. The cheapest and most expensive proposals in a set of three are often both wrong for reasons that have nothing to do with the number.

Sources and change log

Sources. The protocol is our own. The three dated results quoted are from the clawlaw.in programme, recorded in Tier_1/GEO_BASELINE_RESULTS_2026-07-27.md and Tier_1/claude_response_17_09_audit.md, and all three concern public pages and listings, so a reader can check them. No pricing figure on this page comes from any third party survey, because no pricing figure appears on this page.

Change log. 29 September 2026: first published with the full protocol, the six pricing models, the thirteen capture fields, and the statement that zero price pages have been reviewed. When the survey runs, this section will carry the frame size, the number of pages checked, the number that qualified, the capture window, the archive links and the median with its sample size, and this entry will remain above it.

Common questions

Why not publish a range from public sources to be useful?

Because it would not be useful, it would be repeated. A range with no sample behind it gets copied into other pages, and within a year it is a fact with no origin. We would rather publish the method and the zero.

Should our agency publish prices, given that competitors do not?

Publish bands against a defined scope, with a date and what is included and excluded. You will not lose the deals you want, and you become the page an assistant can quote when a buyer asks what this costs. On 17 September 2026 Claude explicitly named price transparency as part of why it ordered vendors as it did.

What if our prices change every quarter?

Then date the page every quarter and keep a short change log on it. A price page with a date and a history is more credible than a price page that pretends to be timeless, and the history is itself the sort of detail that gets quoted.

Is a "starting from" price good enough?

Not by the definition used here, because it has no unit and no scope, so nobody can compare it. If you must use it, attach the scope: starting from a figure, per month, for a named set of deliverables, with a named minimum term.

How long does this survey take to run properly?

Building a frame of two hundred agencies and checking each for a price page is two to three days of careful work, and most of that time is spent deciding whether something is a price. That is why step two is four hard conditions rather than a judgement.

Can we use an assistant to collect the prices?

Use it to help build the frame, never to supply figures. In our own September 2026 audit the assistant withdrew its own claim to have searched in three separate batches, saying it could not honestly substantiate it. A figure with no URL and no capture date is not a data point, whoever produced it.

What to do first

If you are an agency, publish one price band today, with the scope, the date, and what it excludes, and keep the change log on the page. If you are a buyer, ask each shortlisted agency for a written price against one identical scope you wrote yourself, and compare the answers rather than the brochures.