Short answer: a pricing page for AI search is one an assistant can read without running scripts and quote without guessing. That means every plan's name, amount, currency, billing period, pricing unit, inclusions, limits and an "as of" date, written as server-rendered text rather than drawn by a toggle or a calculator. When a buyer asks an assistant what your product costs, the assistant does one of three things: quotes your page, quotes someone else's copy of it, or tells the buyer to contact sales. Which of the three happens is mostly decided by how your pricing page is built. Among the 23 AI visibility vendors we profile, 19 publish an entry price in US dollars, and the ones that do not, or that publish it in a form a crawler cannot read, leave the answer to whoever wrote about them last.

This article explains how AI reads pricing pages, shows the real patterns vendors in one software category use to publish (or withhold) prices, and gives a working template for a plan block that survives extraction. The vendor data comes from the competitor profiles we maintain for our comparison pages, each checked against the vendor's own pricing page between 2026-09-07 and 2026-09-09. We describe the patterns neutrally; the point is the mechanism, not a ranking of anyone's pricing strategy.

  • The crawler sees the HTML, not the page you see. Most AI crawlers do not execute JavaScript, so prices drawn by a script, a monthly/annual toggle or a slider may not exist for them.
  • An unqualified number is a liability. "$99" without "per month, billed annually, per domain" gets quoted as if it were the whole story.
  • Quote-only is a choice with a cost: the assistant cannot place you in any "under $X" answer, and third-party estimates fill the gap.
  • Structured data helps only if it agrees with the visible page, and it cannot express every pricing model.
  • Dates matter. A visible "prices as of" date lets an assistant, and a buyer, judge whether a figure is current.

Why a pricing page for AI search matters more than it used to

Price questions are among the most concrete things buyers ask assistants: "how much does X cost", "is X cheaper than Y", "what is the cheapest tool for Z under $100 a month". In a search results page, the buyer clicked through to your pricing page and read it. In an assistant, the answer is assembled for them, and your pricing page is one input among several. If it is readable and unambiguous, it tends to be the input that wins, because it is the primary source. If it is not, the assistant uses what it can find: a review site's summary, a comparison article from last year, a forum post, or its own memory.

The effect is uneven. A product with a clear published price can be included in answers that filter by budget. A product without one is either left out of those answers or described with a hedge. OpenAI's documentation for shopping in ChatGPT gives a concrete example of budget-driven selection: when a shopper states a budget of $30, ChatGPT focuses more on price. Software and service purchases are not shopping cards, but the same logic applies to any answer that has to compare costs: a price the model cannot find is a price it cannot compare.

Three outcomes when a buyer asks about your price

What the assistant doesWhen it happensWhat the buyer hearsRisk to you
Quotes your pagePrices are readable text with clear conditionsYour number, your conditions, your plan namesLow, if the page is current
Quotes a third-party copyYour page is unreadable, ambiguous, or older copies are more widely repeatedSomeone else's summary, often outdated or missing conditionsHigh: wrong prices and wrong plan names spread
Refers the buyer to salesNo price exists anywhere it can find"Pricing is not public; contact the vendor"Excluded from budget-filtered shortlists

The middle row is the one most teams never see, because nothing in their analytics reports it. A buyer who is told a competitor's "starting price" and an outdated figure for yours does not click through to correct the record; they move on.

How AI reads pricing pages: fetch, extract, reconcile, answer

No vendor publishes the exact pipeline its assistant uses to answer a price question, but the stages any such system must go through are clear enough to design for. The figure shows the four stages and what tends to be lost at each.

1FetchA crawler or user-triggered fetch getsthe HTML. Most AIcrawlers do not runJavaScript.2ExtractPlan names, amounts,currency, period,unit (per seat, perbrand, per credit),limits, overages.3ReconcileCompare with reviewsites, comparisonpages, older copiesand the model’sprior beliefs.4AnswerQuote a number withits conditions, givea range, or say“contact sales”.WHAT GETS LOSTPrices drawn by ascript or a toggleWHAT GETS LOST“Save 20%” withno figure to computeWHAT GETS LOSTOld third-partycopies win tiesIf nothing usablesurvives, the answerfalls back to othersSources: InsightWonder feature rules (JavaScript); Vercel, “The rise of the AI crawler” (2024-12-17).Steps 2–4 are a model of the process, not a vendor specification.
Figure 1. The four stages between a pricing page and a quoted price, and what tends to be lost at each. Source: InsightWonder feature rules; Vercel (2024-12-17). The stages are a model of the process, not a vendor specification.

Fetch: what reaches the system

A crawler building a search index, or a fetcher acting on a user's request, retrieves your pricing page's HTML. Vercel's analysis of AI crawler traffic on its network found that the major AI crawlers it studied, including OpenAI's, Anthropic's, Meta's, ByteDance's and Perplexity's, do not render JavaScript; ChatGPT's and Claude's crawlers fetch JavaScript files but do not execute them. Gemini, which uses Googlebot's infrastructure, and AppleBot do render. Our own rule of thumb is the same: most AI crawlers do not execute JavaScript, and a page whose content appears only after client-side rendering is an empty shell to them.

Pricing pages are unusually exposed to this, because they are often the most interactive page on a marketing site. Common patterns that leave a non-rendering crawler with nothing:

  • A monthly/annual toggle where the default state is filled in by script after load.
  • A seat or usage slider whose output price is computed in the browser.
  • Prices fetched from a billing API at runtime, so the HTML carries placeholders.
  • Currency chosen by the visitor's IP address, so a crawler sees a currency that buyers in your main market never see.
  • A consent or region modal that must be dismissed before the content renders.

The test is simple: fetch the page without JavaScript and search the raw HTML for your prices. Our guide to JavaScript rendering and AI walks through doing that for any page.

A five-minute test for your own pricing page

Run this before you change anything, and again after every redesign of the pricing page. It needs only a terminal.

# 1. What a non-rendering crawler receives (no JavaScript, no cookies)
curl -sL -A "Mozilla/5.0 (compatible; OAI-SearchBot/1.0)" https://example.com/pricing \
  | sed -e 's/<[^>]*>/ /g' | tr -s ' ' | grep -iE 'month|year|seat|USD|\$' | head -40

# 2. Does the currency change with the visitor's location?
curl -sL -H "Accept-Language: de-DE" https://example.com/pricing | grep -oE '(US\$|USD|EUR|€)' | sort | uniq -c

# 3. Is there a date?
curl -sL https://example.com/pricing | grep -iE 'as of|updated|effective' | head

Read the output as a crawler would. Can you name every plan, its price, its currency, its billing period and its unit from the text alone? Is the annual price there as well as the monthly one, or only whichever the toggle shows by default? Does the currency change when the request looks like it comes from another market? If any answer is no, that gap is what an assistant will fill from somewhere else. Note that some sites serve different content to declared bots, so treat this as a first check and confirm with a real fetch from the assistant where you can, for example by asking it to read your pricing page and summarise the plans.

Extract: turning a page into facts

Once the text is available, the system has to turn a visual layout into statements: this plan costs this amount, in this currency, per this unit, for this period, including these limits. Pricing pages are written for eyes, and eyes fill gaps a parser cannot. A column header that says "Annual" two hundred pixels above a price, a footnote marker that links to "billed yearly" at the bottom of the page, or a unit shown only as an icon all become ambiguous once the layout is flattened into text. The more of the condition that sits in the same sentence or line as the number, the more likely it survives.

Reconcile: your page against everyone else's

Your pricing page is not the only source the assistant considers. Review sites, comparison articles, community threads and older snapshots of your own page all carry prices, and they often disagree with each other. Where the primary source is clear and dated, it has an advantage. Where it is ambiguous or missing, the repeated third-party figure tends to win simply because it appears more often and states its number plainly.

Answer: with conditions, as a range, or not at all

The final answer can carry a number with its conditions ("from $X per month per domain, billed annually"), a hedged range, or a referral to sales. You influence which by what you give the earlier stages. Assistants are generally careful with prices, so an unambiguous primary source with a date gives them permission to be specific.

What 23 AI visibility vendors' pricing pages actually show

To ground this in real practice, we looked at how the vendors in one software category publish their prices. The category is AI visibility and generative engine optimisation tools, which we track for our comparison pages. Each profile records the vendor's pricing model, its published entry price, whether a free plan or trial exists, and notes on how the pricing page is presented, all checked on the vendor's own site between 2026-09-07 and 2026-09-09. Prices change; treat these as a dated snapshot of publishing patterns rather than current price quotes.

VENDORS BY PRICING MODEL (23 AI VISIBILITY TOOLS, PRICING PAGES CHECKED 2026-09-07 TO 09-09)Per brand ordomain, tiered14Usage orcredits6Custom quote(demo-gated)2Not published(sales-quoted)1Source: InsightWonder competitor profiles (CMS), each checked on the vendor’s own pricing page. Muted bars: no price amodel can quote. Classification is ours; a vendor with tiers billed per brand counts as “per brand”.
Figure 2. Pricing models of 23 AI visibility vendors as published on their own pricing pages. Source: InsightWonder competitor profiles (CMS), checked 2026-09-07 to 2026-09-09; classification is ours.

Most vendors publish a model a buyer, or an assistant, can reason about: 14 price per brand or domain in tiers, and 6 price by usage or credits. Two sell on custom quotes behind a demo, and one publishes no pricing at any tier. Of the 23, 17 offer a trial, 2 a free plan, 2 neither, and for 2 we could not establish it from their sites.

Three “$99” entry prices, three different purchasesWhat an assistant repeats: “from $99 a month”Profound Starter$99BillingBilled yearlyEnginesChatGPT onlyIncluded50 promptsSemrush AI toolkit$99/moBillingPer domain, billed annuallySeats1 user; extra from $45/moIncluded25 custom promptsSE Visible Basic$99Projects3 projectsIncluded200 promptsThe same headline number buys a different amount of work at each vendor. Only a price stated nextto its billing basis and inclusions can be compared fairly.
Figure 3. Three vendors whose published entry price is $99, side by side with the terms that come with it. Source: InsightWonder competitor profiles (CMS), checked 2026-09-07 to 2026-09-09.

Among the 19 vendors that publish an entry price in US dollars, the range runs from $20 to $800 a month, with a visible cluster around $99. Two details in that range matter for how assistants quote prices. First, the basis is not uniform: Profound's $99 Starter is billed yearly, Scrunch's $250 is the annual-billing rate ($300 month to month), and Semrush's AI toolkit is $99 a month per domain billed annually. An assistant that reports "$99" for all of them is technically quoting each vendor's own page and still misleading the buyer. Second, an entry price can exclude the thing most buyers are comparing: Ahrefs Brand Radar's custom prompts start around $50 a month, while its pre-collected AI visibility index is a separate subscription several times that. A pricing page for AI search has to put those conditions next to the number, because the number will travel without the rest of the page.

Entry prices hide most of the differences

The notes in the same profiles show how much an entry price leaves out. Profound's $99 Starter covers ChatGPT only with 50 prompts, and the full engine list needs higher tiers. Semrush's $99 includes 25 custom prompts and one user, with extra users from $45 a month. SE Visible's $99 Basic includes 200 prompts and 3 projects. Trakkr includes 50 prompts per brand at both its $100 and $500 tiers, so prompt depth does not scale with price until Enterprise. Knowatoa starts at $59 a month with citation tracking, API and multi-market gated to its $199 tier. Evertune's $800 entry includes 100,000 prompts across 11 models. Four vendors at the same headline price can be selling very different amounts of work.

That matters for assistants because the headline number is the part most likely to be repeated, and the inclusions are the part most likely to be dropped. A buyer who asks "which AI visibility tool is cheapest" and hears "$20" or "$29" has learned almost nothing without the prompt allowance and engine list that come with it. The vendors whose pages state inclusions in numbers, in the same block as the price, are the ones whose prices can be compared fairly.

Trials, free plans and seats

Seventeen of the 23 vendors offer a trial and two offer a free plan. Here too, the details are what make a statement quotable: Scrunch's Starter trial runs seven days without a credit card, ZipTie.ai offers a 7-day, 25-prompt trial that needs no card, AI Peekaboo pairs a $29 floor with a 14-day trial, and SE Visible lists a ten-day free trial. Seat pricing is the other frequent condition: Scrunch's Starter includes three seats with extra seats at $25 a month, while OtterlyAI includes unlimited team members. A pricing page that states trial length, card requirement and seat terms as text answers three of the questions buyers most often ask an assistant after "how much".

Five ways to publish pricing, and what an assistant can say about each

The vendor data falls into five recognisable patterns, ordered from most to least quotable. The examples are neutral illustrations from the profiles, not judgements about whether each vendor's choice is right for its market.

AN ASSISTANT CAN QUOTE ITAN ASSISTANT CAN ONLY GUESS OR REFEREvery tierlistedAmounts, units andlimits in the HTML,including the toptiere.g. Rankscale(all tiers)Entry listed,top quotedA real floor, then“contact us” aboveite.g. Similarweb(standalone tier)Listed, buthard to readNumbers drawn byJavaScript, ordiscounts with nofiguree.g. Peec AI, SEVisibleListed with acatchOne currency only,limited-time rate,or a floor thatexcludes the coree.g. Nightwatch,LLMrefs, AhrefsQuote-onlyNo public number;demo or sales callfirste.g. Brandlight,Bluefish, ConductorSource: InsightWonder competitor profiles (CMS), vendor pricing pages checked 2026-09-07 to 09-09.Examples are neutral illustrations of each pattern, not ratings of the vendors.
Figure 4. Five ways vendors publish pricing, ordered by how much an assistant can quote, with neutral examples from the same vendor data. Source: InsightWonder competitor profiles (CMS).

Every tier listed

Rankscale publishes a price for every tier including Enterprise, which our profile notes is rare in this category. An assistant can answer almost any price question from such a page, including "what does the top tier cost". The catch our profile records is a different one: its credits make the real cost depend on engine count and cadence rather than the tier alone, and its $20 Essentials tier is priced publicly while its contents are not detailed. A complete price list still needs the inclusions to be equally explicit.

Entry listed, top quoted

Similarweb's AI Search Intelligence publishes a standalone entry price, unusual for an enterprise suite, and packaging above that tier is quote-only. This is the most common compromise in business software: it gives assistants a floor they can cite ("from $X") and tells them truthfully that larger deployments are priced individually. It works well when the page states what triggers the move to a quote (seats, brands, regions, contract terms), so the assistant can tell a buyer which side of the line they are on.

Listed, but hard to read

Two different problems fall here. Peec AI's public pricing page renders no numbers without JavaScript; the $95, $245 and $495 tiers in our profile come from Peec's own page of instructions for AI systems. That second page is an instructive workaround: the vendor knows its main page is opaque to crawlers and publishes the facts separately in plain text. SE Visible shows annual pricing only as "save 20%" without exact figures. Neither is hidden pricing, but both force an assistant to compute or guess, and models are rightly cautious about arithmetic on prices they cannot verify.

Listed with a catch

Nightwatch publishes prices in euros only, with no US dollar tier, so a US buyer carries the exchange rate and an assistant answering in dollars has to convert or decline. LLMrefs flags its single tier as limited-time pricing, so the rate may not hold, which is honest and also gives an assistant reason to hedge. Ahrefs, as above, publishes a starting price that our profile describes as ambiguous by design. In each case the number is real, but it needs its condition attached to be quoted correctly.

Quote-only

Brandlight and Bluefish publish no pricing, offer no self-serve signup and no trial: everything is demo-gated. Conductor publishes no pricing at any tier; everything is sales-quoted. For an assistant, these produce the same answer: pricing is not public. That is accurate, but it also means these products cannot appear in any answer that filters by budget, and any figure a buyer does hear will have come from a third party.

PatternWhat an assistant can truthfully sayMain riskCheapest fix
Every tier listedExact prices for each planInclusions or usage costs unclearState inclusions and overage prices in numbers
Entry listed, top quoted"From $X; larger plans priced individually"Buyer cannot tell which side they fall onName the thresholds that trigger a quote
Listed, hard to readPossibly nothing, or a third-party figureCrawler sees placeholders; discounts cannot be computedServer-render default prices; print both billing periods
Listed with a catchA number with a hedgeThe hedge is dropped when the number is repeatedPut the condition in the same line as the number
Quote-only"Pricing is not public"Excluded from budget answers; third-party guesses fill the gapPublish a floor, a range or the pricing drivers

Usage and credit pricing: the hardest model for AI to quote

Six of the 23 vendors price by usage or credits. Usage pricing is fair to buyers and hard to describe in one sentence, which makes it the model most often misquoted. Our profiles note the problem from the buyer's side: AthenaHQ's credit-based consumption makes monthly cost harder to predict, and Rankscale's credits make the real cost depend on engine count and cadence rather than the tier alone. ZipTie.ai takes the opposite approach and publishes usage-based pricing to the cent: you choose engines and cadence and see the price change. That is excellent for a human, and only quotable by an assistant if the resulting prices also exist as text, not only as the output of an interactive calculator.

Publish worked scenarios, not only a formula

An assistant can repeat a sentence; it is much less reliable at running your pricing formula. The most quotable way to publish usage pricing is a short list of worked scenarios in plain text, each with its inputs and its result: a small team at the low end, a typical customer in the middle, and a heavy user at the top. With placeholders in place of real figures, the shape looks like this:

ScenarioInputs (stated in text)Monthly cost (stated in text)
Small team[n] prompts, [n] engines, weekly refreshUS$[amount]
Typical customer[n] prompts, [n] engines, daily refreshUS$[amount]
Heavy user[n] prompts, all engines, daily refreshUS$[amount], or contact sales above [n]

Alongside the scenarios, state the unit price of a credit, what one credit buys (one prompt on one engine, for example), and what happens when credits run out. Those three facts let an assistant answer "how much would it cost me" for a buyer who describes their situation, which is the form most real price questions take.

Keep the calculator, add the text

None of this means removing an interactive calculator. Calculators are useful for buyers who know their numbers. The requirement is only that the calculator is not the sole place the prices exist. Server-render its default state with real values, and keep the worked scenarios and unit prices in the HTML below it.

Quote-only pricing: when it is the right call, and how to make it quotable anyway

Quote-only pricing is not a mistake. It fits products whose cost genuinely depends on scope (enterprise suites, services, regulated industries), sellers who negotiate every deal, and markets where a public price would anchor buyers too low. The question for a pricing page for AI search is not whether to publish every number, but whether the page gives an assistant enough to say something useful and true.

Decision criteria

  • Does a meaningful share of buyers self-qualify on price? If many buyers screen by budget before talking to anyone, invisibility in budget-filtered answers costs real pipeline.
  • Is there a real floor? If no deal closes below a certain amount, publishing that floor costs little and qualifies buyers for you.
  • Are third parties already publishing figures? If review sites or comparison articles estimate your price, those estimates are what assistants repeat. Publishing your own replaces a guess with a fact.
  • How often does pricing change? Frequent changes favour fewer, more stable public numbers (a floor, a range) over a full price list that will go stale.
Question an assistant may be askedQuote-only pagePage with a published floor and drivers
"How much does X cost?""Pricing is not public""From $X per year; priced by number of brands and regions"
"Tools like X under $500 a month?"Usually omittedIncluded or excluded correctly
"Is X more expensive than Y?"Third-party estimate, or no answerA comparison grounded in both vendors' pages
"What drives X's price?"Generic guessThe drivers you listed, in your words

Making a quote-only page quotable

If you keep quote-only pricing, give the page the facts an assistant can use without a number: who the product is priced for, what the price depends on (seats, brands, markets, volume, contract length), the minimum contract term, what is included in every plan, whether there is a trial or pilot, and how long a quote takes. Each of those is a sentence an assistant can repeat accurately. "Pricing is annual, based on the number of brands and markets tracked; every plan includes API access and onboarding; pilots are available" is a far better answer for a buyer than "contact sales", and it is entirely in your control.

A worked example: a quote-only page, rewritten

Consider a typical enterprise pricing page, written here as an illustration rather than taken from any vendor. It shows three plan names, each with a "Talk to sales" button, a feature matrix of ticks, and a line saying "Flexible pricing for teams of every size". To a crawler, that page says three plan names exist and nothing else. An assistant asked about cost can only say pricing is not public, and if a review site once estimated a figure, that figure becomes the answer.

The rewrite keeps quote-only pricing and changes what the text says. Under the plan names, one paragraph states that pricing is annual and depends on the number of brands and markets tracked; that every plan includes API access, onboarding and a named contact; that the minimum term is twelve months; that a paid pilot is available; and that quotes are sent within a stated number of business days. A second paragraph says who each plan is designed for. Nothing confidential is published, no number a competitor could undercut appears, and the assistant now has five accurate sentences to give a buyer instead of one unhelpful one. If the company is willing to publish one more fact, "contracts typically start from US$[amount] per year" turns the page from unquotable to quotable without exposing the price list.

Building a pricing page for AI search: the anatomy of a quotable plan

The figure below is a template for one plan block. Every line is a fact a buyer needs to compare plans, and every line is written as text that will survive being fetched without scripts and flattened into a paragraph.

TeamUS$[amount] per brand / monthBilled monthly; annual: US$[amount]/moPer brand, unlimited seatsIncludes: [n] prompts, [n] enginesOver the limit: US$[amount] per [n]For: teams tracking one brandPrices as of [date]. Taxes extra.1Plan name, as text2Amount with explicit currency3Billing period, both prices4The unit you are charged by5What is included, in numbers6Limits and overage price7Who the plan is for8A visible “as of” dateIllustrative template: bracketed values are placeholders, not real prices. Every line is server-rendered text, not an image or a toggle.
Figure 5. A plan block for a pricing page for AI search, annotated with the eight facts an assistant needs. Illustrative template; bracketed values are placeholders.

The eight elements

  1. Plan name as text. Not only in an image or an icon. Use the same name everywhere, including invoices and help articles, so an assistant can match it.
  2. Amount with an explicit currency. "US$" or "USD", not only "$", if you sell internationally. If you show local currencies, keep one canonical currency in the HTML regardless of visitor location.
  3. Billing period, with both prices if you offer both. Print the monthly and the annual-billing price as text. A toggle can stay for humans, but its default state must not be the only price in the HTML.
  4. The pricing unit. Per seat, per brand, per domain, per project, per thousand credits. This is the condition most often lost when a number is repeated.
  5. What is included, in numbers. Seats, projects, prompts, engines, storage, whatever the plan meters. "Generous limits" cannot be compared.
  6. Limits and overage prices. What happens at the limit: a hard stop, an overage rate, or an upgrade. Usage-priced products need this most.
  7. Who the plan is for. One line. It helps an assistant match a plan to a buyer's situation, which is how most price questions are actually phrased.
  8. A visible "prices as of" date. It tells an assistant and a buyer how current the figures are. Our guide to content freshness and dates covers how to show dates without faking freshness.

Before and after: one plan block rewritten

The rewrite below uses placeholders rather than real prices, because the point is the structure. Notice that nothing is hidden in the "before" version; the information is there for a human looking at the rendered page, and still not there for a system reading the text.

ElementBefore (written for eyes)After (written for extraction)
PriceLarge "[amount]" with a toggle above it"US$[amount] per brand per month, billed monthly; US$[amount] per month billed annually"
UnitA small icon of a person"Per brand. Unlimited seats."
Inclusions"Everything in Starter, plus more""[n] prompts, [n] engines, daily refresh, CSV export"
LimitsFootnote marker linking to the bottom of the page"Over [n] prompts: US$[amount] per additional [n]"
AudienceNone"For in-house teams tracking one brand in one market"
DateNone"Prices as of [date]. Taxes extra."

The same principle runs through all writing for answer engines: put the condition in the same sentence as the claim. Our guide to extractable writing covers it beyond pricing.

Comparison tables on the pricing page

Most pricing pages end with a feature matrix of ticks and crosses. Keep it, but do not rely on it. A tick in a cell is meaningless once the table is flattened and separated from its column header. For the handful of features buyers actually choose on, state them in text next to each plan as well, and use words rather than symbols in the table cells ("Included", "Not included", "Add-on, US$[amount] per month").

Pricing page structured data: what helps and what does not

Structured data can state prices in a machine-readable form, and for some page types search engines document exactly what they expect. Google's software app structured data, for example, lists the app's name, offers.price and a rating or review as required for rich-result eligibility; a free app sets offers.price to 0, and for paid apps Google recommends also including offers.priceCurrency. Google also says it does not guarantee that features consuming structured data will show.

For a pricing page, schema.org's Offer type with a price, a priceCurrency and, where needed, a UnitPriceSpecification for the billing period and unit can describe simple tiered plans. A minimal example for one plan, with placeholder values:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "SoftwareApplication",
  "name": "ExampleApp",
  "applicationCategory": "BusinessApplication",
  "offers": [{
    "@type": "Offer",
    "name": "Team",
    "price": "99.00",
    "priceCurrency": "USD",
    "priceSpecification": {
      "@type": "UnitPriceSpecification",
      "price": "99.00",
      "priceCurrency": "USD",
      "unitText": "per brand per month, billed monthly"
    }
  }],
  "aggregateRating": { "@type": "AggregateRating", "ratingValue": "4.6", "reviewCount": "128" }
}
</script>

Three cautions. First, structured data must agree with the visible page; a JSON-LD price that differs from the displayed one is a contradiction an assistant may resolve either way, and Google treats mismatches as a quality problem. Second, Google's software app guidance requires a rating or review, and those must be genuine and visible to users; do not add an aggregateRating to make the markup validate. Third, structured data cannot express everything: credit-based pricing, overage tiers and "contact us above N seats" have no clean representation. Use markup for the simple facts and keep the conditions in the text. Our article on schema markup for AI search covers which types earn their keep.

Keeping a pricing page for AI search, and the answers built on it, current

A correct pricing page is necessary but not sufficient, because assistants also draw on what others have written about your prices and on older copies of your own page. Keeping answers current is a maintenance routine, not a one-off fix.

On your own site

  • Keep one pricing URL. If you must move it, redirect the old URL permanently rather than leaving an orphan with old prices.
  • Change the date when you change prices. And only then; a date that changes without a price change teaches readers to ignore it.
  • Retire old plan names explicitly. A short note ("the Pro plan was renamed Team in [month, year]") helps an assistant reconcile old third-party references with the current page.
  • Update help-centre articles and documentation that mention prices or limits at the same time, so your own site does not contradict itself.
  • Consider a plain-text summary for AI systems. Peec AI's separate page of AI instructions carrying its tier prices is one real example of a vendor doing this. If you use llms.txt or a similar file, keep its prices in sync with the pricing page or leave prices out of it.

Off your site

Review platforms, marketplaces and comparison articles carry their own copies of your prices. You cannot control them, but you can supply corrections where platforms accept vendor updates, and you can make sure your own page is the most authoritative and most current statement available. Where a widely read article quotes an old price, a polite correction with a link to your dated pricing page is usually welcomed by editors who care about accuracy.

A quarterly pricing-page review

Pricing pages drift. A plan gets renamed in the product but not on the page, a limit changes in the billing system but not in the feature matrix, a regional page keeps last year's prices. A short quarterly review catches most of it:

  1. Fetch the pricing page without JavaScript and confirm every plan's name, price, currency, period and unit appear as text.
  2. Compare the page against the billing system's current price list, line by line, including overage rates.
  3. Check that help-centre articles, documentation and any plain-text summary for AI systems carry the same numbers.
  4. Check that structured data on the page matches the visible prices, or remove it.
  5. Search for your product's price on the main review and comparison sites in your category and note any outdated figures.
  6. Ask the assistants your buyers use what your product costs, and record the answer and the source they cite.
  7. Update the "as of" date only if something changed.

The review takes an hour for most companies. Its value is that it turns pricing accuracy in AI answers from something you discover when a prospect quotes a wrong number on a sales call into something you check on a schedule.

Measuring whether assistants quote you correctly

Write down the price questions buyers actually ask: "how much does [product] cost", "[product] vs [competitor] pricing", "cheapest [category] tool for [situation]", "does [product] have a free plan". Run them across the assistants your buyers use, on a schedule, and record three things for each answer: whether a price is given, whether it matches your current page including its conditions, and which source is cited. The pattern over time tells you whether your page is winning the reconcile step. InsightWonder's prompt monitoring runs a fixed question set like this across engines and keeps the cited sources, so a wrong price can be traced back to the page it came from.

Sources