ScalePost Review 2026: It Calls Our Method a Guess, and It Is Half Right

ScalePost’s homepage describes our method and calls it a guess. The sentence is: “Other tools ask AI a question and write down the answer. That’s a sample. That’s a guess. ScalePost sits at the CDN layer and watches what AI actually fetches.”

Every study we publish is prompt sampling. So this review has to do something the rest of our catalogue does not: take the criticism seriously, say which half of it is correct, and then show precisely where the alternative instrument is blind.

The short answer is that they are right that a sample is a sample, and wrong that a fetch is a citation.

Disclosure: EchoWi sells prompt-based AI visibility measurement, which is the method this vendor argues against, so read this knowing we have a position. Every quotation below was read from ScalePost’s own homepage on 7 August 2026. Our own figures come from our measurement register, with prompts, markets, dates and run counts published.

The short version

1. No published price. There is no pricing page; /pricing, /brands, /publishers and /methodology all return 404.
2. The data source is real and different: first-party CDN logs via Fastly, Cloudflare and Akamai.
3. Their criticism of prompt sampling is correct, and we have published the numbers that prove it.
4. Their own step three does the thing they say they never do. A CDN log records a fetch, not a citation.
5. Both instruments have a blind spot. They answer different questions and the review says which.

What it is, verified

ScalePost sells to two sides. To publishers: “Your content is already being cited by AI. ScalePost tells you exactly how much, by brand and by URL, so you can charge for it.” To brands: “Find the publishers actually driving your AI visibility.”

The method is described in four steps on the homepage, under the heading “How the measurement works. Four steps. No sampling. Every number traceable to source.”

The sample dashboard shows beauty and skincare over the last 30 days across ChatGPT, Perplexity, Gemini and Claude: “6,824,193 actual AI citations measured at the publisher CDN this period. No prompts. No sampling.” Underneath, a ranking of publishers by citations and share of brand mentions, led by elle.com at 2.8M and 34.6%.

Where they are right, with our own numbers

The criticism deserves a straight answer, because we can quantify it from our own published work rather than concede it vaguely.

A single prompt run is a draw, not a rate. We ran the same question about running shoes three times on Google’s AI Overview and got five domains, all five in every run: perfect reproducibility. We ran a question about GEO tools in the United States and got 16 domains with 5 stable one day, and 23 domains with 3 stable on another. Same question, same market, same surface.

How much repetition a category needs varies enormously. Our stability study put the range at roughly three runs to roughly thirty, depending on the category. Anyone quoting a percentage off three runs in a volatile category is quoting noise.

And one surface is not the web. We asked two questions of four surfaces at the same moment and got 51 distinct domains, none of which appeared on all four. Forty-one appeared on exactly one.

So yes: a prompt-based measurement is a sample, its precision depends on how many runs you buy, and it describes one surface at one moment. We publish that limitation on every study because it is true. ScalePost is pointing at a real weakness and most of this market does hide it.

Where the CDN instrument is blind

Now the other half, and it is a specific technical claim rather than a rhetorical one.

Steps 01 and 02 describe fetches. Step 03 calls them citations.

A CDN log records that a bot requested a URL from a server. It does not record:

• whether the fetched page was used in an answer,
• whether an answer was produced at all,
• whether a human ever saw it,
• or which brand, if any, that answer named.

Getting from a logged fetch to “tagged to the URL, the brand mentioned, and the AI surface it came from” requires modelling, matching and assumption. That is inference. The same paragraph says “Nothing inferred”, and the sentence above it says “Real URLs, real counts, nothing inferred”. Both cannot be true. The counts are real; the interpretation of a fetch as a citation of a named brand is not a count, it is a conclusion drawn from one.

This matters for the headline number. Retrieval systems fetch far more than they cite: they crawl for indexing, they re-fetch for freshness, they grab candidates and discard most of them. 6.8M is entirely plausible as fetches and implausible as answers a person read. Without a stated ratio between the two, that figure cannot be compared with anything.

Three more limits follow from the architecture, none of which is a flaw so much as a boundary:

• It only sees publishers who integrated. “Top publishers driving AI visibility for Sephora” is a ranking inside ScalePost’s panel. A publisher outside it contributes zero by construction, which is exactly the failure mode of every panel-based measurement in media history.
• It cannot see an answer that cites nobody in the panel, including answers that cite your competitor’s own site, a forum, or a video. In our own data, YouTube reached eleven of twelve measurements and Reddit six. Neither is a premium publisher with a CDN deal.
• It cannot answer the buyer’s actual question. “Am I in the answer when someone asks which tool to buy, in Spain, in Spanish, this week” is a question about answers. A fetch log cannot express it.

The two instruments, side by side

Neither is the truth. One takes a photograph of the answer and has to repeat it to know what is stable. The other counts a real event that is one step removed from the thing you care about.

ScalePost’s own page half concedes this. Further down it proposes a stack: “ScalePost, CDN signals, actual AI citations + Google Analytics, human traffic, conversions + Prompt tools, sample answers, directional = Complete AI visibility.” That is a more honest slide than the one at the top of the page, and it is close to the position we would defend.

Where ScalePost is the right call

• You are a premium publisher trying to price your AI influence. This is the strongest use of the product by some distance, and nothing in our catalogue competes with it. Turning bot fetches into an advertiser-facing report is a genuinely new commercial argument.
• You buy media and want to know which publishers to fund. “Which titles does AI read about my category” is a question a fetch log can answer better than a prompt sample.
• You need an audit trail. “Audit-ready. Sellable. Defensible.” is fair for counts of logged events, whatever they are called.

Where it is the wrong call

• You want to know whether AI recommends you. That is a question about answers, in a market, in a language.
• Your category is not covered by premium publishers. B2B software, professional services and most local categories are answered from vendor pages, forums and video.
• You need a price to plan. There isn’t one published.
• You need to compare it with anything. Its unit is not the unit anyone else reports, and the fetch-to-citation ratio is not stated.

What we could not verify

• Any price. No pricing page exists on the site.
• How many publishers are in the panel, or what share of any category they represent.
• How a fetch becomes an attributed citation with a named brand. The mechanism is asserted, not described.
• The fetch-to-citation ratio, without which 6.8M cannot be interpreted.
• Whether “1,700+ bots” includes surfaces that answer users, or is mostly crawlers, scrapers and agents that never produce a consumer-visible answer.
• Whether the sample dashboard is real data or an illustration. It names a real brand and real publishers with precise figures, and the page does not say which it is.

ScalePost versus the alternatives

The full verified table is in our comparison of GEO tools.

The useful thing here is not the verdict, it is the question it forces. Every other tool in this catalogue argues about engine counts and prompt allowances, which are differences of degree. ScalePost argues about what should be counted at all. That is the more interesting fight, and the correct response to it is not to defend prompt sampling as if it had no limits. It is to publish the limits, which we do, and to hold the other instrument to the same standard, which is why “nothing inferred” cannot survive its own step three.

Common Questions About ScalePost and CDN-Based Measurement

How much does ScalePost cost?

No price is published. There is no pricing page on the site, and /pricing, /brands, /publishers and /methodology all return 404 as of 7 August 2026. The only call to action is “Get started”.

Is CDN data better than prompt-based AI visibility tracking?

It is different, and better for a different question. CDN logs count real fetch events by AI agents on publishers who integrated, which is the right instrument for a publisher pricing its influence or a media buyer choosing titles. Prompt sampling observes the answer itself, in any market and language, including sources no panel covers. Neither is a census of AI visibility.

Does a CDN log prove your brand was cited in an AI answer?

Not on its own. A log entry records that a bot requested a URL. Whether the page was used, whether an answer was produced, whether anyone saw it, and which brand it named are not in the log. Reaching those requires inference, which is worth knowing when a page says “Nothing inferred”.

What is wrong with prompt-based measurement?

It is a sample. Ask the same question twice and the source list can change, and how much repetition a category needs runs from about three runs to about thirty. One surface is also not the web: we asked two questions of four surfaces at once and 41 of the 51 domains cited appeared on only one. Any vendor quoting a precise percentage without a run count is selling you noise.

Which method should I buy?

Ask what you will do with the answer. If you are selling advertising against your AI influence, count fetches. If you are deciding whether to write a page, change your pricing table or enter a market, you need to see the answer itself, repeated enough times to tell stable from accidental. Most brands need the second and some need both, which is roughly what ScalePost’s own stack slide proposes further down the same page.

Does EchoWi have a conflict of interest in this review?

Yes, and it is stated at the top. We sell prompt-based measurement and this vendor argues against it. That is why this review concedes the specific criticism with our own published figures instead of dismissing it, and why the objection we raise is a technical one about the step where a fetch becomes a citation rather than a complaint about tone.