Field Notes #1: What AI reads on Webflow sites (and gets wrong)

This week we checked 5 content-heavy Webflow sites: 2 B2B software companies and 3 professional services firms, from 463 to 2,934 pages each. Three things stood out:

• Pricing: both software companies had a pricing answer in ChatGPT, Gemini or Claude that didn't match their own pricing page.
• Old pages: all 5 sites still publish pages about 2024 or earlier: old news, past events and expired tax guidance. AI crawlers read these pages like any other.
• Placeholder text: 2 of the 5 had template text such as "Lorem ipsum" left in the page code. Visitors never see it, but the crawlers do.

None of this is about how good the content is. It's what happens when a site grows to hundreds of pages and nobody owns what's already published. AI tools then repeat whatever they find, old or new.

1. AI tools quote prices your pricing page doesn't show

We asked ChatGPT, Gemini and Claude the questions buyers ask, such as "How much does [company] cost?", and compared each answer with the company's own pages. For both software companies, at least one engine gave a price that didn't match the pricing page.

• A discount that doesn't exist. Two of the three engines described a cheaper rate for one customer segment. That rate appears nowhere on the company's pricing page.
• Pages that disagree with each other. One homepage says plans start at one price, while the pricing page starts lower. An engine quoted the lower annual figure. The AI wasn't really wrong: the site gave two answers, and the AI picked one.

The second case is the more common pattern. When your own pages disagree, AI tools pick one at random, and that version gets repeated to every buyer who asks. Pricing changes are the usual cause: the pricing page is updated, but the homepage, comparison pages and old blog posts aren't.

Small sample: 2 sites, 6 questions each, 3 engines. We'll report this every week as the sample grows.

2. Every site still publishes pages nobody has reviewed in years

All 5 sites have pages whose address or title names 2024 or earlier: 205 pages in total. Most are harmless on their own: a 2018 tax-year article, an announcement for a 2022 webinar, an award shortlist from 2021. Together they're a problem, because AI tools can't tell an archive page from current advice.

We then looked closely at up to 10 of these pages on each site. On one site, 7 of 10 should come out of search results (noindex), and on another, half do. On the other three, they mostly need updating. Many had no visible date, no named author, and no sign that they're historical. That's exactly what makes an old page look current to an AI crawler.

The accounting firm is the clearest case: 63 of its old pages are tax guidance or news, including articles about filing for tax years 2017 and 2018. To an AI tool, that's still guidance from an accounting firm.

3. Placeholder text hiding in the page code

On one site, 101 of the 110 pages we checked carry "Lorem ipsum" placeholder text in their page code. It comes from a template element that was styled out of sight but never deleted. Visitors don't see it. AI crawlers read the raw page, so for them every article ends with a paragraph of Latin filler. On a second site, release notes still contain Webflow's default "This is some text inside of a div block".

It's an easy mistake on Webflow, where hiding an element is a click and deleting it from a CMS template is something nobody gets round to. The fix is usually one template edit, and it cleans up every page built from that template at once.

A quick check on your own site: open one of your blog posts, view the page source (Ctrl+U / Cmd+Option+U) and search for "lorem" and "div block".

What to fix this week

1. Make your prices say the same thing everywhere. Search your site for every price, plan name and "starting from" figure. Then ask ChatGPT, Gemini and Claude what you cost, and compare. Fix the pages, not the AI.
2. Decide what your old pages are for. For anything about a past year, choose one: update it if the topic is still relevant, merge it into a current page that covers the same topic, noindex it if it's only worth keeping as an archive, or remove it if it's of no use to anyone.
3. Clear out template leftovers. Check one page from each CMS collection for placeholder text, empty headings and missing titles.

How we checked

• Old pages: we read each site's full sitemap and flagged pages whose address or title names 2024 or earlier. We ignored sitemap dates, which Webflow resets whenever a site is published.
• A closer look: up to 130 pages per site, checking visible dates, markup and the raw page code, plus a full audit of 10 old pages per site.
• Prices and facts: we asked ChatGPT, Gemini and Claude up to 6 buyer questions per company. We only counted an answer as wrong when we could quote both the AI's answer and the company's own page.
• Privacy: all companies are anonymised.

Field Notes is a weekly look at what AI reads on content-heavy websites, published every Sunday. Next week: a bigger sample, and what caused the wrong answers we found.