Technical report

How we read a pricing page, and the three ways it goes wrong

Three reading rounds across 18 vendors. Eight could be compared at all. Two of those eight moved an annual rate while leaving the headline monthly price untouched, and the third round overturned a verdict the second had reached.

Every figure on this site has a date on it, and this is the report that explains what that date is worth. A script reads each vendor’s pricing page, stores what it saw with the timestamp of the reading, and reports what moved since last time. On October 30, 2026 it reported six vendors as changed. Opening the snapshots, two of those six had not changed price at all: the reading had changed. A third round on November 15 then overturned one of the verdicts the second round had reached. That is the finding, and it is the reason nothing reaches a comparison page on the strength of a log line.

Software pricing moves quietly. A vendor renames a tier, lowers a limit or withdraws annual billing, and nothing announces it – the page simply says something different than it did last quarter. Comparison articles are then wrong without anyone noticing, including ours. We keep a dated record so that ours can be checked, and the data is published openly so that anyone else can check it too.

What follows is the method, three rounds of results, and the four failure modes in full. The failure modes are the useful part. A collection method whose limits are undocumented is worse than no method, because it produces confident wrong answers at scale. The fourth was added after it produced one.

The method

It is deliberately simple, because every piece of cleverness added to a scraper is a new way for it to be wrong without saying so.

  1. Fetch the page over HTTP. One request per vendor, to the pricing URL recorded for that vendor – the same URL the comparison pages link to, so the reading and the published figure cannot drift apart.
  2. Extract the lines carrying currency figures, together with whatever label the page puts on the same line. The result is a list of price lines, not a parsed price model.
  3. Record the reading with its metadata: source URL, fetch timestamp in UTC, HTTP status, and which currency symbols appeared.
  4. Diff against the previous reading for that vendor and append any difference to a change log.
  5. Open the live page before any figure is published. This step is manual and it is the one that matters. The script produces leads. A person produces findings.

Two things it deliberately does not do. It does not convert currencies, so a figure is recorded in the currency it was advertised in. And it does not normalise price lines into plan names and billing periods, because doing that automatically is where the errors come from – a monthly price and an annual-billed-monthly price look identical as numbers and mean different things.

What three rounds produced

Readings on and .

  • 18 vendors configured
  • 15 produced at least one readable snapshot
  • 3 produced none at all – LowFruits, Mangools and WordRocket
  • 27 snapshots: 4 vendors have three readings, 4 have two, and 7 have one
  • 373 rows in the deposited dataset – 340 price lines and 33 discount lines

Coverage is uneven because vendors were added to the tracker over time rather than all at the start. Seven vendors still have a single reading, which means there is nothing to compare them against and no change could have been detected for them in any round.

That settles what the second round’s headline figure meant. Six vendors had two readings and the change log flagged six vendors, which is the same six: every vendor that could be compared was reported as changed. Six of eighteen would suggest a third of the market repriced in six weeks. Six of six says something different and less interesting about vendors – that a text diff over a marketing page flags nearly everything, and that the work is in reading what it flagged.

The third round is the first one that could check the second. Eight vendors now have more than one reading and four have three, so for the first time a conclusion drawn from two readings could be tested against a third. One of them did not survive.

Round two: the six flagged on October 30, adjudicated

Each of these came out of the change log as "CHANGED". What the snapshots show is not the same thing in every case.

Ahrefs

Artefact, concealing a real change

Every figure appears to move because the page was read in euros in September and dollars in October. Ahrefs sets currency by the reader's location, so this records where the script ran. Inside that noise, the October reading carries two price lines the September one does not.

Keysearch

Artefact

Logged as $24 and $48 removed. Both figures are present in both readings. September rendered each plan's price twice in the server HTML and October rendered it once, so the diff reported the lost duplicates as removals.

Search Atlas

Real, and the largest of the round

Eleven price lines gone, including every annual figure, replaced by a single line reading "$199/mo billed monthly". A withdrawal of annual billing is a real change and the one this round was worth running for.

Distribb

Real, on its face

$97 and $495 replaced by $79 and $399, with two annual totals appearing that were not there before. Same currency in both readings, so nothing about the method explains it away.

Baby Love Growth

Needs a browser reading

Four euro figures replaced by four lower euro figures. This vendor draws its plan cards in the browser and has quoted both euros and dollars across readings, which is the combination that makes a server-HTML diff least trustworthy.

Outrank

Needs a browser reading

$199/month removed, $1899/month and $99/month added. A tenfold figure appearing is more often a new table on the page than a repricing of an existing plan, and the server HTML cannot tell you which.

Two of six confirmed as artefacts, two real on their face, two still open. We are not calling the last two either way from server HTML, which is the whole discipline this report is describing. Both will be read in a browser before any figure from them is published, and if that reading contradicts what is written here, this report gets corrected and says so.

Round three, and the verdict it overturned

Readings on November 15, 2026 for the five vendors due, plus SEO Writing on the 17th. Six vendors differ from their previous reading. The entry that matters is the first one, because it is about this report rather than about a vendor.

Distribb

Round two was wrong, and only a third reading could show it

Round two called this "real, on its face": $97 and $495 replaced by $79 and $399, same currency both times, nothing about the method to explain it away. The November reading carries $97 and $495 and not the lower pair. Read all three rounds together and the stored page text settles it. September: $97 /month and $97 /month Billed yearly. October: $97 beside $79 /month Billed yearly ($948). November: $97 /month and no yearly line at all. The two October figures were two billing bases, not a before and an after. The monthly price never moved. What appeared in October was an annual discount, and what is absent in November is the annual line itself.

Baby Love Growth

Closed, and the headline price was never the story

Round two left this open pending a browser reading. The October snapshot carries · save 40% and · save 50%; the November one carries neither, and states the saving in cash instead – · save $480 and · save $1,916 against $99 and $399 a month. Both work out at 40%. The Scale discount was cut from 50% to 40%, the annual rate rose 20%, and the monthly price did not move. The full working, and why two readings in different currencies could still answer it, is under the second failure mode below.

Moz Pro

A discount change with no price behind it

All 27 price lines identical between September 15 and November 15. The only difference is one discount line: Last chance to save up to 40% - ends today! is gone, while Pay yearly (Save 20%), the standing plan discount, is unchanged. An expiring banner and a change to what a plan costs both read as "a discount moved", which is why the two are reported as separate lines and neither is read as the other.

SEO Writing

A promotion with a deadline, held until it passes

The page restructured between readings and now carries 40% off yearly plans, every year. and Get 40% off, which the October reading did not. The yearly figures moved with it: $168, $708, $1,428 and $2,760 are presented as plan prices rather than as a 25% saving off a monthly rate. A promotion with a countdown on it is not a list price until the countdown passes, so this one is held until December 7, 2026 and nothing from it has been published.

Ahrefs

One line, and the currency held still this time

Read in dollars in both October and November, so the currency switch that made round two’s Ahrefs diff unreadable is absent. Twenty price lines became nineteen: $100/mo per user is no longer present. A single seat-price line disappearing from server HTML is as likely to be a page restructure as a change to what a seat costs, so it is recorded and not published.

Search Atlas

Still unreadable, and the two new lines are not plans

One price line in October, three in November, and both additions are credit top-up rates rather than plan prices. No plan price is legible in server HTML in either reading. The annual withdrawal this vendor was flagged for in round two is not restored in the November reading, which is consistent with that verdict without independently confirming it.

Two of the eight comparable vendors moved an annual rate and left the monthly price alone. Distribb and Baby Love Growth are unrelated companies on unrelated pages, and in both cases the figure a buyer watches did not move. That is the most useful thing in this report for anyone comparing these tools: watching the headline monthly price is not watching the price.

And a verdict drawn from two readings did not survive a third. Round two had Distribb as a genuine reprice and said in terms that nothing about the method explained it away. Something did. Two figures in one reading were two billing bases, and that stood as a published conclusion for sixteen days. It is now the fourth failure mode below, because the three we had did not cover it.

The three failure modes

1. The page has no prices in it

The script reads server HTML and does not run JavaScript. Several vendors in this category build their plan cards in the browser, so the reading comes back empty or partial. Three of the eighteen produced no readable snapshot at all.

The dangerous version of this is not the empty reading. It is the half-empty one. An empty file is obviously a failure. A file with some of the prices in it looks like a successful reading of a page that has fewer prices than it used to, which is indistinguishable from a vendor withdrawing a plan. The rule that follows: an empty or partial reading means the script could not read the page. It never means the vendor removed its prices.

2. The diff reports a text change as a price change

The clearest worked example in the dataset is Keysearch on October 30, 2026. The change log reports two removals: removed: $24 and removed: $48. Read as pricing, a budget tool just dropped both of its plans.

Both figures are present in both readings. In September the server HTML rendered each plan's price twice – the price lines read $24, $24, $48, $48 – and in October it rendered each once. Starter was $24 and Pro was $48 on both dates. The diff reported the lost duplicates as removals.

What actually changed is how the page builds its billing toggle, which is a fact about the page and not about the price. A text diff cannot tell those apart, and it is not reasonable to ask it to. The only defence is opening the snapshot, which takes under a minute and is the reason this one never reached a published page.

3. The currency moved, so everything looks like it moved -- including the vendor

Ahrefs sets currency by the reader's location. The September reading is in euros and the October reading is in dollars, so all 18 of September's price lines differ from October's: €46.7/mo against $50/mo, €179/mo against $199/mo, €37.4 per extra user against $40. The plans did not change. The script moved.

It would be easy to draw the wrong lesson here and write a rule that a wholesale currency switch is always a location change and can be ignored. That rule would have cost us a real finding. The October reading carries two price lines the September one does not – a bare $99/mo and a bare $299/mo, taking the page from 18 price lines to 20. Those two sat inside 38 lines of diff noise, and the log recorded the whole thing as a single change.

So a currency switch does not only manufacture false positives. It conceals true ones. The right response is not a rule but a reading: when the currency moves, the diff is worthless and the two snapshots have to be compared line by line by a person.

Ahrefs is the easy version of this, because the switch sits between two readings by one tracker and announces itself across every line at once. Baby Love Growth is the hard version, and we read it wrong for two months. Its plan cards are built in the browser, so the tracker reads nothing at all and every figure comes from a person opening the page. Four such readings – September 13, 2026, the 15th, September 30 and November 2 – alternated between dollars and euros while the monthly numbers stayed 99 and 399 throughout. We recorded that as the vendor changing its currency four times in two months, and wrote it into our own notes as a thing that vendor does.

It is not something that vendor does. On November 15, 2026 the same page, read twice on the same day, served $99 and $399 to one browser and €99 and €399 to another request. Same page, same numbers, two symbols. Every flip in that list was a different request, not a vendor decision. The pattern we had recorded as vendor behaviour was a record of our own readings.

Two things follow, and the first is the one that changes what a price means. No single currency symbol on a page is correct for every reader, so a figure has to carry who saw it as well as when. Our own comparison of Keysearch and Baby Love Growth says $99 where our other pages say €99, and both are true – for different readers. Neither is an error and neither needs correcting. The second is that we first wrote this up as the currency depending on where the reader sits, and that went further than the evidence: the dollar reading came from a browser in Europe with no VPN, and the euro reading from a German network address. You cannot read the reader's location off the symbol, which is the same shape of inference as the one this section exists to warn about, made one step along. What we can say is that the symbol is a fact about the request and not about the vendor's price list.

Which leaves the question these two readings were supposed to answer, and a way to answer it. That vendor's Scale plan read 199 a month billed yearly on October 30, 2026, beside the words "Pay yearly, save 50%". On November 15 both readings, the dollar one and the euro one, showed a yearly saving of 1,916 against 399 a month: 40%, working out at 239. A percentage carries no currency symbol, so the discount on that plan was cut from 50% to 40% whichever reading you take, the annual price rose 20%, and the monthly price did not move at all. The other plan's discount is unchanged. So two readings in two currencies cannot be subtracted, but the ratios inside each one can still be compared – and here that is the only thing that caught a 20% increase a tracker watching the headline figure would have scored as no change.

4. Two figures in one reading may be two billing bases

The script extracts price lines. It does not know which of them are alternatives to each other. A plan card showing $97 /month beside $79 /month billed yearly produces two price lines, and so does a plan whose price was cut from $97 to $79. In the stored reading those two situations are identical.

This is the near relative of the Keysearch artefact in the first round, where two identical figures in one reading turned out to be one price rendered twice. The general form is worth stating on its own: a reading contains figures, and a price comparison is made of the relationships between them. The relationship is the part that is not in the file.

The defence is not a cleverer parser. It is the surrounding page text, which every snapshot stores in full for exactly this reason, and a second billing basis recorded for every plan rather than one number. This failure produced a wrong published verdict and it stood for sixteen days, which is why it is written down here rather than fixed quietly.

What we do about it

  • A logged change is a lead. Nothing reaches a comparison page from the change log. The live page is opened first, and the date on the page is the date of that reading.
  • An unreadable page is recorded as unreadable, never as a page whose prices were removed.
  • Currency is recorded as seen and never converted. A figure converted between currencies and then presented as a price is an invented number.
  • A currency difference between two readings is a fact about the two requests, not about the vendor. We spent two months recording one vendor's currency as moving between readings. It was not moving. We were. And the symbol does not tell you who the reader was: we guessed a country from a dollar sign and were wrong about that too.
  • Both billing bases get recorded where a vendor publishes both, because a monthly price and an annual-billed-monthly price are different numbers and vendors lead with whichever is lower.
  • Corrections are published as corrections. We have run the wrong figure on this site and said so on the page where it ran. A method that never produces a visible correction is not a method anyone is checking.
  • Both billing bases are recorded, always. A plan has a monthly price and an annual price, they move independently, and two of the eight comparable vendors in this dataset moved one without the other. A single number per plan cannot represent that, and a comparison that mixes the two silently reverses its own verdict.

What we could not confirm

  • Whether Outrank genuinely repriced. It was flagged on October 30, has not been read since, and has no third-round reading. It is listed as open, not as a change. Baby Love Growth, flagged in the same round and left open in the same way, was closed by the November 15 readings and is written up above.
  • Whether the three unreadable vendors moved their prices at all during this period. We have no reading for LowFruits, Mangools or WordRocket in either round, so for them this dataset is evidence of nothing.
  • What anyone actually paid. This is a record of advertised list prices. Discounts, legacy plans, negotiated rates and enterprise pricing behind a sales call are all outside it.
  • Whether a vendor's page said something different to a different reader on the same day. Ahrefs demonstrably varies by location. Others may vary in ways one reading cannot detect.
  • Whether two identical figures in a reading are two prices or one price rendered twice. The Keysearch case was resolved by reading the surrounding page text. There is no general rule, and a dataset user should treat duplicate figures within a single reading with suspicion.

The data, and how to check this

The dataset behind this report is deposited openly and is not reproduced here. It holds the whole thing as one CSV, the individual readings with their source URL and fetch timestamp, the change log, and the list of pricing pages read.

Kloth, R. (2026). AI SEO and AEO Tool Pricing Snapshots[Data set]. Zenodo. https://doi.org/10.5281/zenodo.23964371Concept DOI, which always resolves to the newest version. Licensed CC-BY 4.0. One honest note on that record: it is served with a noindex robots tag because the depositing account is not on an institutional email domain, so it is citable and resolvable but invisible to search engines. That is Zenodo's behaviour for unverified accounts and not something we can change.

If you want to check a specific figure rather than the method, price history lists what has moved and when, and methodology covers what this site does and refuses to claim across everything else it publishes.

Questions

Does RankJudge test the AI SEO tools it compares?

No. RankJudge runs no benchmarks and claims none. Every figure on this site is something a vendor published on its own page on a stated date, or published research by a named organisation. This report documents how those pages are read and where the reading fails.

Can a price tracker report a price change that never happened?

Yes, and it did. The change log is a text diff between two readings of the same page, so it reports what moved in the text rather than what moved in the pricing. On October 30, 2026 it reported Keysearch's $24 and $48 as removed. Both figures are present in both readings: September's HTML rendered each price twice and October's rendered each once, so the lost duplicates read as removals. Starter was $24 and Pro was $48 on both dates. The snapshot is the source and the log is only a lead.

Which AI SEO tools do not show their prices in server HTML?

LowFruits, Mangools and WordRocket, in this round. The collection script fetches server HTML and does not run JavaScript, so a vendor that builds its plan cards in the browser returns a page with no prices on it at all – WordRocket's is about 6.6KB and carries neither the prices nor the plan names. Of 18 vendors configured, those three produced no usable snapshot. An empty reading means the script could not read the page, never that the vendor removed its prices.

Why does Ahrefs show different prices in euros and dollars?

Ahrefs sets the currency from the reader's location rather than from an account setting. It was read in euros on September 15, 2026 and in dollars on October 30, 2026, so all 18 of its price lines differed between the two readings while the plans themselves were unchanged. Each figure in this dataset records the currency the page displayed where the script ran. None has been converted from one currency into another, and none should be treated as though it had been.

Can a currency switch hide a real price change?

Yes, and that is the more dangerous direction. The October Ahrefs reading contains two price lines, $99/mo and $299/mo, that the September reading does not. Because the currency switch made all 18 lines differ, those two sat inside 38 lines of noise and the log recorded the whole thing as one undifferentiated change. A rule that says to dismiss currency switches would have dismissed them too.

How often do AI SEO tools change their prices?

This dataset cannot tell you, and the honest answer is worth more here than a figure. Only six of the 18 vendors had two readings, so only six could be compared at all, and the change log flagged every one of them. Two of those six were artefacts of how the page was read rather than price movements, and two more have not yet been read in a browser, which leaves two confirmed changes. A repricing rate built on that would be a number with nothing behind it.

Where can I download the AI SEO tool pricing data?

The pricing snapshots behind this report are deposited openly under DOI 10.5281/zenodo.23964371, licensed CC-BY 4.0, and are not duplicated on this site. The deposit holds the full dataset as one CSV, the individual readings with their source URL and fetch timestamp, the change log, and the list of pricing pages read.

Nobody publishes their failure modes, because a method with four named ways of being wrong reads weaker than one with none. It is the other way round. Two of six flags in one round were artefacts, a verdict from that same round did not survive the next one, and the only reason those are sentences in a report rather than wrong prices on a comparison page is that somebody opened the files. Treat any pricing comparison that shows you figures without a date, a source and an account of what its collection misses as a comparison that has not checked.

This report is also available as a PDF, which is the version deposited for citation. The page is the canonical copy: if the two ever disagree, the page is right and the PDF is stale.