Profound sells AI visibility. In June, AI could not read its pricing page.

Profound tells companies what AI engines see when they look at them. On 15 September it announced a $180 million Series D at a $1.8 billion valuation, less than seven months after a $96 million Series C. Sequoia and Kleiner Perkins led it. The company says revenue tripled in six months and it has more than 1,000 enterprise customers, including Comcast, Walmart and The Estée Lauder Companies.

By late June, less than three months before that round, the company had learned that bots could not see the plans on its own pricing page.

The story comes from research Kyle Poyar ran with Nikolas Laskaris, a researcher at Profound. Profound was building a tool to check whether AI crawlers could read a given page, and tested it on its own site first. To a person the pricing page looked normal. To a bot, the self-service plans were missing. An outside post about Profound, published around the same time, found the same gap independently.

You can see it in the Internet Archive. The Wayback Machine's copy of tryprofound.com/pricing from 23 June 2026 holds the navigation, the footer and not much else. It has about a thousand characters of readable text, no plan names, no prices. JavaScript drew the plans in the browser. Most AI crawlers run little or none of it, so they got the empty frame.

Profound moved its pricing to server-side rendering and shipped the fix on 25 June. The archived copy from 15 July has the prices in the raw HTML: Starter at $99 a month, Growth at $399, both billed yearly. Profound says citation-bot traffic rose 13% week over week after the change and kept climbing. The pricing page became its second most-cited page on the whole site.

Those two numbers are Profound's own, and Profound is a sponsor of the newsletter that published them. Treat them as a company's report, not an audit. The archive copies are not self-reported, and they show the before and after plainly.

HubSpot had the same bug, and had to work around the page

Profound's problem is common. In the same research, Poyar and Laskaris checked the public pricing pages of the Cloud 100, a list of top private cloud companies. 57 of 77 were fully readable to bots. 10 hid at least 40% of their body text from them.

HubSpot, a public company and so not on that list, was caught out too. Aja Frost, a former marketing leader there who is now Director of Growth Marketing at Mercury, described it in her account of HubSpot's twelve months of AI-search experiments. A team member noticed that LLMs were quoting the wrong prices. Sales call transcripts showed it happened fairly often. The cause was the one Profound found: HubSpot's pricing pages rendered in JavaScript. Googlebot runs JavaScript, so this had not been much of a problem for search. AI bots, Frost writes, "can't (yet)", so they took prices from outdated third-party pages.

Frost could not simply fix the page. Another team owned it. So her team went around it and wrote plain blog posts, one per product. Each described that product's pricing in full and was structured for bots. HubSpot's server logs showed bots crawled the blog heavily. Over two months, AI pricing accuracy improved for five of HubSpot's six products.

The sixth, Sales Hub, got worse. The team fixed it by correcting the outdated Sales Hub prices on the third-party sites the models were reading. Frost calls that time-intensive but reliable.

HubSpot's pricing page is still built the same way for people. We fetched hubspot.com/pricing/marketing on 26 September with an ordinary browser user-agent. The HTML that came back had no dollar figures in it, only a JavaScript shell of about 36 characters of readable text. The same URL fetched as GPTBot, ClaudeBot, PerplexityBot or Googlebot returned a different, fully rendered page. It had roughly 29,000 characters of text, the Starter, Professional and Enterprise tiers, and about a dozen distinct dollar figures. HubSpot now gives crawlers a prerendered copy and leaves the browser version alone. We cannot say when that started.

The question after the page is readable

Once bots can read the page, there is still the question of whether the engines go there at all.

Poyar and Laskaris tested 7,600 AI answers to pricing prompts across six engines, collected 29 July to 10 August 2026. Company pricing pages were rarely the main source. Three outside sites came up again and again: Vendr in 18.7% of runs, Reddit in 18.6%, G2 in 15.9%.

For the companies that did well, one good pricing page was not enough. They had several first-party pages that agreed with each other. Plaid was the clearest case. Its pricing page was cited first 42% of the time. Across answers the engines cited its billing documentation in 70%, its pricing page in 64% and its FAQs in 50%. The docs carried the most weight because they hold the detail: billing models by product and endpoint, edge cases, and a plain statement of which numbers are not public.

Companies with no public price got the opposite. One vertical AI unicorn with no pricing page drew negative pricing language in 67 of 76 answers. The engines, often reading Reddit threads, put its price at "roughly $100 to $2,400 per user per month."

What Profound's page says now

Profound's pricing page changed again this month.

The last Wayback copy with the $99 and $399 plans is from 14 September, the day before the Series D. The live page we fetched on 26 September is server-rendered, and its brand plans are easy for a bot to read. It lists two plans for brands: a free Trial and Enterprise, priced "Custom." Starter and Growth are gone. A GEO tools comparison from Morningscore also reports that Profound dropped its self-serve brand plans in mid-September, with agencies on a separate self-serve Agency Growth plan. That agency plan was already on the page in July.

That is a normal move for a company that now counts more than 1,000 enterprise customers. But it puts Profound on the other side of its own research. For brands, the page is readable, and there is no price on it to read.

The only dollar price left on the page is Agency Growth, at $99 a month. It appears only after a visitor clicks "For agencies," and JavaScript draws it in the browser. It is not in the HTML a crawler fetches.

The old brand prices have not gone away. When we asked Tavily's extraction API for the same URL on 26 September, it returned a clean, healthy-looking copy of the page with $99 and $399 on it, from a cache. Review and comparison sites still list those prices as current. For now, a machine asked what Profound costs has two sources to choose from: Profound's page, with a free trial and an Enterprise button, and a stack of third-party pages quoting plans that no longer exist.

Run the test on your own site

Four checks cover most of it. The first two take a minute each.

1. Fetch your pricing page the way a bot does. curl it, or load it with JavaScript turned off, and read what comes back. If the prices are not in that HTML, most AI crawlers do not see them. Poyar and Laskaris name four usual causes: client-side rendering, content behind tabs or sliders, robots.txt and other blocks (some inherited and never cleaned up), and pricing widgets loaded in an iframe from another domain.
2. Fetch it again as a crawler. Send a GPTBot or ClaudeBot user-agent. If the answer differs from the browser version, as HubSpot's does, know which copy each audience gets and keep them in step.
3. If you do not own the page, publish around it. HubSpot's route: one plain post per product, stating the price in full, on a part of the site bots already crawl. Then correct the stale prices on the third-party pages the engines quote.
4. Write the billing docs, not only the page. Plaid's docs were cited more often than its pricing page. If a number is not public, say so in your own words, or the engines will fill the gap from Reddit.

If you take prices off the page, as Profound just did for brands, the cached copies of your old prices are what the machines can still find and quote. Update or retire them on purpose.