We re-priced 21 AI search tools — our categories were already wrong

Finding

What we set out to check, and what we found instead

Our categories were wrong before we started the re-price. Six of the seven tools we had filed as pure SEO or content tools were already selling AI-search visibility on the day we built the original sheet. We did not notice until the Internet Archive proved it against us.

We should say the commercial interest plainly: we build and sell AI-visibility and content tooling in the same category we are writing about. The procurement sheet we are correcting here is the one we used internally to argue that our own stack is cheaper than the competition. Five of its errors appear in this article. Four of them flattered us.

Here is what we expected to find. On 25 May 2026 we logged prices and category labels for 21 tools. One tool from the original 22 was excluded, because it serves a non-English market outside our scope. We planned to return after roughly two months and document price drift. The working assumption was that a fast-moving category would show meaningful movement: new tiers, discontinued plans, repositioned entry points.

On 21 July 2026, 57 days later, we ran the re-check. Fifteen prices came directly from published HTML. Five required loading the page in a browser to render. One tool published no price at all. The prices had barely moved.

That was the expected finding, and it was the less interesting one.

What we had not planned to test was whether our original category labels were accurate. We tested them because several rows read oddly next to the live pages: a tool we had filed as a writing suite was leading with citation tracking. So we pulled archived versions of each vendor's homepage and pricing page from around the 25 May baseline.

That result was uncomfortable. Six of the seven tools we had filed under SEO or content were already using AI-search framing on their own pages at baseline. Phrases describing answer-engine visibility and AI citation tracking appeared in their copy on the day we wrote the sheet. We had missed it. The Internet Archive had not.

This matters for how the rest of the article should be read. The convergence we are describing did not happen between May and July 2026. It had already happened. Our sheet reflected a category split: GEO tools on one side, SEO tools on the other. These vendors had already abandoned it in their own copy. Buying guides published in 2026 still print that split.

We are correcting the sheet here because it is in active use. We are correcting it in public because the errors that flattered us are the ones most likely to mislead someone else.

Method

How this was measured

We fetched every pricing page on 21 July 2026 using a TLS-fingerprinted HTTP client that runs no JavaScript. That matters: the fetcher reads only what the server sends in the initial HTML response. If the price appeared in that served text, we recorded it as read from HTML. Fifteen of the twenty-one pages met that bar.

For the remaining pages, a person opened each URL in a full browser and read the price from the rendered output. Five pages required that step. One of those five returned no body to our fetcher at all — not a redirect, not an error page with content, just nothing ¹. One vendor publishes no price anywhere we could find ².

That gives us 21 tools in the table. We tracked 22 in total; the twenty-second was a Russian-market text-quality tool, excluded as out of scope for an English-market comparison.

The archive cross-check, and why we ran it

Fetch-and-read answered the pricing question cleanly enough. It did not answer the category question, and that is where we went wrong first.

Our original plan was straightforward. Compare how each vendor described its product in May against how it described it in July, then flag anyone who had added AI-search framing in between. Seven vendors looked like they had shifted. We were ready to write that up as evidence of the category converging in front of us.

Then we ran the archive check. For each of those seven vendors we pulled the Internet Archive snapshot nearest to the 25 May baseline. Then we searched it for the same AI-search phrases we had recorded as new in July.

Six of the seven phrases were already there in May.

That broke the first conclusion entirely. Those phrases had not appeared between May and July. They had been present at the baseline date and we had simply not recorded them. What our baseline captured was not the vendors' own words — it was our classification of those words. It was wrong when we wrote it.

What the flaw actually was

We built the baseline from our own procurement sheet. It is a working document, used to track tools we are evaluating or already paying for. That sheet recorded our read of each vendor's positioning, not a verbatim copy of the page. A sheet like that is a summary, and a summary inherits the assumptions of whoever wrote it. Ours inherited a two-category model of the market that the vendors had already left.

An archive check is now a standing step in how we run this kind of comparison. Without it, a procurement sheet makes a poor baseline: it records the analyst's categories, not the source.

Archive evidence

Six of the seven were already there

We went back to the archives expecting to document a wave of repositioning. Most of it had already happened before we ran our original sheet.

Six vendors in our set had already moved their public-facing copy toward AI-search visibility by the time we captured our baseline. The seventh, Semrush, shows the same framing today. But the Internet Archive holds no capture of its pricing page inside our window, so we do not count it as a confirmed change. That leaves six. Here is what the archives show, vendor by vendor.

Surfer

The archived page from 23 May 2026 carries the title Surfer: AI Visibility Platform ³. The hero already reads Boost visibility in Google Gemini ChatGPT Perplexity and beyond. A banner on the same snapshot promotes an in-person event on how brands win visibility in AI search. None of that is new copy. Today the same site reads Surfer - AI Visibility Platform for maximum organic growth ⁴.

Our sheet recorded Surfer as “SEO content optimization, Business $219.” That framing was already out of date on the day we wrote it.

One detail is worth separating out. In May and again in July, one URL carried two identities. The older page title Surfer | SEO Content Optimization Platform still sits beside the newer AI-visibility hero. Nobody removed the old title. We cannot say whether that is a metadata lag or a deliberate hedge. Either way, anyone reading a browser tab or a search snippet still sees the legacy framing while the hero points somewhere else.

Writesonic

The 25 May 2026 archive shows the title Writesonic | The AI Search Growth Engine. Win Customers. and a hero reading Win customers from AI search. See where AI ignores you. ⁵ We checked it again in July. Title identical. Hero identical, word for word ⁶.

Our sheet called Writesonic an “AI writing tool, Standard seats.” The archived page was already something else.

Ahrefs

The 26 May 2026 snapshot of the Ahrefs pricing page already carries it: helping your business stay discoverable in search, AI, and beyond ⁷. Today's page carries the identical sentence ⁸. Nothing changed between then and now, because it was already there.

Our sheet said “SEO suite, Lite $129.” That price is accurate. The category label missed what the page was already saying.

Clearscope

This is where our sheet was most plainly wrong. We recorded Clearscope's Essentials tier at $170 a month. The archived pricing page from 5 May 2026 — three weeks before our baseline — already reads Simple, transparent pricing for SEO and AI search visibility. It already prices Essentials at $129 a month ⁹. Not $170. $129. The current page shows the same sentence and the same price ¹⁰.

We were wrong about the price on the day we wrote it, and we missed the framing at the same time.

Frase

The 24 May 2026 archive shows the Frase product menu already listing GEO and AI Visibility as named entries ¹¹. July has moved further. The tagline is now The content operating system for AI search. The menu reads GEO Optimization — Tune content to get cited by AI and AI Visibility — Track where AI search cites you ¹².

So Frase did add language between May and today. But the direction was set before our baseline. Our sheet described Frase as “content ops and brief automation, Team $115” — a description that ignored what the product menu was already advertising.

Jasper

The 26 May 2026 archive shows the Jasper solutions menu already reading SEO, AEO & GEO ¹³. Today the page has sharpened that into GEO & AI Optimization — Win the new front door of search. A banner offers to Optimize your brand for AI search with our GEO Agent ¹⁴.

Again: July is more explicit, but the shift was visible in May. Our sheet said “AI writing, brief to draft to refine.” That described a product the May snapshot had already moved past.

Semrush: excluded from the count

Semrush's pricing page today is titled SEO & AI Search Plans and Pricing. The plan toggle reads SEO + AI Search. The company describes itself as the platform that unifies SEO authority and AI visibility ¹⁵.

We are not counting Semrush among the six, because the Internet Archive holds no capture of its pricing page inside our measurement window. We cannot put an archived quote next to a current quote and draw a line between them. We do not know when the rename happened. We are reporting what the page says today and leaving it there.

Across the six, one pattern repeats. The copy had moved. Our categories had not kept up. In the Clearscope case the gap was three weeks; in the Surfer case it was two days. We recorded what we expected to find rather than what was on the page.

What moved

What actually changed in 57 days

Four changes are documented with archive evidence. One is not. We separate them because the distinction matters when you are deciding how much to trust a comparison sheet — including ours.

GPTZero rewired its billing logic. The archived pricing page from 17 May sells seats. Plans were priced per member, so a team of ten paid for ten slots ¹⁶. Today the same URL sells word quotas — Premium at $12.99 a month, Professional at $24.99 — metered by words scanned rather than headcount ¹⁷. That is not a price change, it is a model change. Compare a per-seat figure from May against a per-word figure today and you are comparing two different products.

Clearscope moved its unit of measurement from topics to prompts. The 5 May archive shows the Essentials tier offering 20 Tracked Topics ⁹. Today the same tier lists 50 Tracked Prompts ¹⁰. Fifty is more than twenty, but the label is the more interesting signal. A topic is a keyword idea a writer tracks. A prompt is a question someone types into a model. One feature, pointed at a different object.

Jasper promoted GEO from a navigation item to a named product surface. In May it sat inside a solutions dropdown alongside SEO and AEO. One entry in a list ¹³. Today it has its own agent and its own pitch ¹⁴. Whether the underlying capability changed is not something public pages can tell us. What changed visibly is where the company puts it in the hierarchy of things it sells.

Frase rewrote its homepage line. The 24 May archive does not carry the phrase The content operating system for AI search ¹¹. Today that line leads the page ¹². One sentence is a small surface, but homepage leads are deliberate.

Sitebulb's price change is the one we cannot fully verify. Our sheet recorded $35 for Sitebulb Pro in May. Today the page shows Lite at $18, Pro at $42 plus $11 per additional user, and Cloud from $125 ¹⁸. There is no archive capture of that page inside our window, so this rests entirely on our two readings. We include it because the gap matters to a buyer, and we flag it so you can weight it accordingly.

Those are the five movements we found. Four have archive support. One does not. Everything else in the sheet held.

Prices

What the prices did

Most prices did not move. Of the 21 tools we tracked, the majority matched the figures we recorded in May. The longer version has three parts. A handful of real shifts. One structural detail in how Otterly bundles engines. And two numbers in our own sheet that are not prices at all.

The moves that actually happened

Sitebulb Pro rose from $35 to $42. Semrush came down to $117.33 on annual billing, against $139.95 for the equivalent plan in May. Screaming Frog reads as $16.58 a month if you divide its published $199 annual licence by twelve ¹⁹. That framing is ours; it is not a number Screaming Frog puts on its page.

Two entry prices came in below our seat maths, because we had priced them for a ten-person team instead of reading the published entry tier. Copy.ai now starts at $29 ²⁰. Anyword sells its Platform tier at $49 a seat ²¹.

The Clearscope row deserves a direct correction rather than a line in a table. Our sheet shows $170 dropping to $129. That looks like a price cut. It is not: the $129 was already live in May when we first pulled the data. We recorded the wrong number at baseline. The move is ours, not theirs.

What did not move

Ahrefs Lite held at $129. Scrunch Plus held at $250. AIVO held across all three tiers ²². Peec's top tier reads $495 against the $499 we recorded ²³. AthenaHQ's entry tier reads $295 against $299 ²⁴. Both differences are small enough to treat as unchanged, and we print the exact figures rather than round them. Across 21 tools, stability was the norm.

How Otterly structures its tiers

Otterly runs Lite at $29, Standard at $189, then $489 and $1,020. One detail matters for buyers: the base plan tracks four engines, and Claude and Gemini are paid add-ons rather than included by default. If your monitoring depends on either model, the effective entry cost is higher than the Lite headline suggests.

This is also the row we are least confident in. Otterly's pricing page returned no body to our fetcher on 21 July, so the figures come from a browser reading rather than an HTML pull.

The two numbers that are estimates

Two figures in our sheet must not be cited as prices, because no published page supports either.

Start with MarketMuse at $149. No page we found supports that number, and we cannot reconstruct where it entered the sheet.

Then there is Profound's agency figure. That was our own construction, assembled from partial signals rather than read from a published price. It may be useful as an order of magnitude. It is not a price Profound has published, and it should not be quoted as one.

Self-audit

Five things our own sheet got wrong

We build and sell AI-visibility and content tooling in this same category. The sheet we are correcting is the one we used to argue that our own stack is cheaper. We found five errors. Four of them flattered us. One worked against our own case. Direction matters as much as the error itself.

Clearscope — we overstated a competitor's price by 32%. We wrote $170 for Essentials. The archived page from 5 May, three weeks before our baseline, already showed $129 ⁹. We were not catching a price change; we were carrying a stale number that predated our own measurement window. Error class: stale capture. Overstating a direct competitor's price made our own offering look cheaper. This flattered us.

MarketMuse — we carried a phantom price for two months. We wrote $149 for Standard. The archived pricing page from 16 May — nine days before our baseline — already showed Free, Optimize, Research and Strategy tiers. Each carried a “Book a demo” button and no price ²⁵. The number was already unsourceable when we wrote it down. Error class: phantom price. This flattered us too.

Originality.ai — we understated a cost we were arguing against. We wrote $14.95 as the team cost. That is one seat; additional seats are $9.95 each, so a ten-writer team pays $104.50 ²⁶. This is the one error that worked against our own case. We were arguing that per-seat models create hidden cost, then under-counted the example that made the point. Error class: unit confusion.

Profound — we built a comparison on a structure we could not source. We wrote an agency base of $399 plus $399 per client workspace. The page today publishes Starter at $99, Growth at $399 a month billed yearly, and Enterprise as tailored pricing. A toggle switches between brand and agency views ²⁷. The per-client ladder we had written down is not on the page. The 21 May archive capture is a JavaScript shell with no prices rendered ²⁸. We cannot date when the structure changed, or confirm it was ever published in that form. Error class: unverifiable architecture. A per-client ladder looks more expensive than a flat tier, so this flattered us as well.

SerpApi — we estimated where a public number existed. We wrote “custom, about $200.” The published tiers are $25, $75, $150 and $275 ²⁹. There was never a need to estimate, and the estimate ran $50 above the tier it should have matched. Error class: unnecessary estimation. Direction: it flattered us.

Five errors, five classes: stale capture, phantom price, unit confusion, unverifiable architecture, unnecessary estimation. Three of them — stale capture, phantom price and unverifiable architecture — are sourcing failures that can appear in any comparison built from periodic manual checks. If you are auditing your own sheet, those are the classes to check first.

Counter-argument

The counter-argument: maybe these are still two categories

The clearest version of the opposing argument comes from a 2026 buying guide by Stackmatix. It states that "GEO tools is a dedicated software category — distinct from traditional SEO platforms — that tracks how AI search engines like ChatGPT, Perplexity, and Google AI Overviews cite and describe" brands and content ³⁰. That is a precise claim and it deserves a precise response.

Where the argument holds. Its underlying logic is sound. Traditional SEO platforms were built to measure ranking signals in a link-graph world: crawl coverage, backlink authority, keyword position in a ten-blue-links result. Tracking how often a brand is cited inside a generated answer is a different measurement problem. Different data model, different feedback loop. A tool purpose-built for that job could reasonably occupy its own category. A session-replay tool sits apart from a web analytics platform, though both watch user behaviour.

Where the evidence pushes back. What we observed is that the boundary, if it was ever clean, is not visible in how these products present themselves. Frase leads with The content operating system for AI search and lists both GEO optimisation and AI-visibility tracking in its product menu ¹². Jasper carries GEO & AI Optimization — Win the new front door of search with a GEO Agent attached ¹⁴. Surfer's own headline is “AI Visibility Platform” ⁴. Writesonic's is “The AI Search Growth Engine” ⁶. Ahrefs sells being discoverable in search and AI ⁸. Clearscope prices “SEO and AI search visibility” as one thing, on one page ¹⁰.

These are not fringe vendors hedging. They are established content and SEO platforms that have absorbed the language into their primary pitch. If GEO were a separate category, we would expect these vendors to stay in their lane, or to announce an expansion. Instead the copy treats AI visibility as continuous with what they already sold.

A third-party signal points the same way from the other side. A 2026 buying guide ranks Otterly, Surfer and Semrush inside one list without separating them into two categories ³¹. That could mean its editors already see one category, or that they collapsed two for editorial convenience. Our evidence does not let us choose between those readings.

The honest limit. Page copy proves what a vendor claims to sell. It does not prove what the product does. A platform can call itself an AI-visibility tool and ship nothing behind it. We ran no feature audits and tested no citation-tracking accuracy against live ChatGPT, Perplexity or Google AI Overviews output. We read what vendors published about themselves.

So the convergence we can demonstrate is convergence of positioning. Whether the products themselves have converged — shared pipelines, overlapping features, interchangeable workflows — is a question this evidence cannot settle. Treat homepage language as a starting hypothesis, not a conclusion.

Buying

What this means if you are buying

Your shortlist question is no longer “GEO tool or SEO tool.” We tracked 21 tools. The boundary had already blurred before we started measuring. Buying against a clean two-column matrix will send you toward the wrong questions on a vendor call.

Start with the framing question, not the feature list. When a vendor's page and your notes disagree, ask to see the feature inside the product. Not on a slide. Six of the seven phrases we had classified as newly added were already present in the May archive snapshots. The disagreement was in our notes, not on their pages. That pattern repeats in buying situations more often than it should.

How stale is a tool comparison?

Our sheet was 57 days old. It carried five pricing or sourcing errors: Clearscope, MarketMuse, Originality.ai, Profound, SerpApi. It also carried six category labels that were wrong on the day they were written. That is roughly one row in four with a bad price, and six of seven with a bad label. Prices held more stable than positioning did, and positioning is what drives the shortlist. If you are working from a comparison published more than two months ago, assume some of the labels are wrong.

Run this check yourself in an afternoon

You do not need a custom fetcher. The method is simpler than ours.

1. Open each shortlisted vendor's pricing page. Note the exact price and the exact positioning sentence.
2. Pull the Internet Archive snapshot of that same page, from the date your comparison was written.
3. Search the snapshot for the phrases your comparison used to classify the tool.
4. Flag every row where the phrase was already there. Those are classification errors, not product changes.

If the phrase was already there at baseline, the pivot your comparison described may be a classification error rather than a product change. Budget about twenty minutes per vendor. For a shortlist of five, that is an afternoon.

The gap between the entry tier and the quote

One vendor in our table publishes no price at all ². Five of the 21 had prices we could confirm only by opening the page in a browser. That asymmetry is a signal. Tools that hide a price tend to have a wide gap between the cheapest published tier and the quoted one. We did not obtain enterprise quotes, so we cannot give you that gap as a figure. What we can say is that if a pricing page needs JavaScript to show a number, budget time for a negotiation rather than a signup.

One practical upshot, and it is narrow. Ask three things before you shortlist. What does the product do today that it did not do twelve months ago? What does the entry tier actually exclude? And is the page you are reading the same page that was live when your comparison was written?

We sell in this category. That does not change the method, but it is worth naming before you weight the conclusions.

Limitations

Limitations

Twenty-one products, one observer, two dates. That is the entire evidentiary base for this article.

Page copy is not product capability. We read pricing and positioning from marketing pages, not from API documentation, contracts or live trials. What a page says a tool does and what it does in a client environment can differ.

Five rows rest on a person reading a page. Surfer, GPTZero, Sitebulb, Peec and Otterly did not expose their prices to our fetcher. One of those pages returned no body to the fetcher at all. Those five carry more interpretation risk than the fifteen we read from HTML.

The archive has gaps we cannot paper over. There is no capture of the Semrush pricing page inside our window. So we cannot date its plan rename, and we do not count it among the six. There is no capture of the Sitebulb pricing page, so its $35-to-$42 move rests on our own two readings. The Profound capture from 21 May is a JavaScript shell with no prices in it, so we cannot date that restructure either. All three vendors still appear in the price table with their current published figures; what we lack is the dated “before” for those specific claims.

Our own sheet introduced five errors into this work. Four flattered us. We corrected all five before publishing, but they existed, they skewed in our favour, and we found them only while preparing this article.

We have a direct commercial interest. We build and sell AI-visibility and content tooling. The sheet we are correcting is the one we used internally to argue our stack is competitively priced. Publishing the errors is a partial correction for that. It is not a complete one, because we cannot fully audit the framing choices we made before we knew the errors were there.

Sources

Sources

All live pages were fetched on 21 July 2026. Archive links are Internet Archive captures; the date in each entry is the capture date.

Related

The work this connects to.