November 2, 2026
Day91 Labs Brief — Issue 016: Research · AI and company facts
We set out to test how often ChatGPT, Gemini and Perplexity get SaaS prices wrong. First we had to write down the right answer, so we read the 24 companies' own pricing pages the way a machine does: as text, without running scripts. Only 12 of 24 stated a clear entry plan, price and billing period. Before anyone can grade an AI's answer, the source has to say something gradable. Often it doesn't.
1. Only 12 of 24 pricing pages state a clear entry price.
Evidence tier A — 24 official pricing pages opened on 31 Oct 2026. Counts are our calculation.
Of 24 B2B SaaS pricing pages, 12 (50%) showed a clear entry plan, price and billing period in their text: Salesforce, monday.com, Asana, Zoom, DocuSign, Shopify, GitLab, Figma, Miro, Airtable, Pipedrive and Webflow. Nine showed a price with something missing or qualified: HubSpot, Dropbox, Wix, Notion, Canva, ClickUp, Calendly, Zapier and Grammarly. Three showed no prices at all, because they load them by script: Jira, Box and Klaviyo. On 12 Oct the same check found 14 clear, 8 qualified and 2 with no prices.
These are not obscure vendors. They are the 24 products buyers are most likely to ask an AI about, and half of them don't say, in plain text, what the cheapest paid plan costs and how it's billed. A person can click a toggle. A model reading the page, or a crawler building the index that AI search relies on, often can't.
Open your own pricing page with scripts turned off, or ask someone to paste its text into a plain document. If the entry price, the unit and the billing period aren't all there in words, assume AI assistants are guessing at them.
Sources
2. The most common gap: a price with no billing period.
Evidence tier A — Same 24 pages. Grouping of reasons is our calculation.
Five pages show a price but never say, in the text, whether it is the annual or the monthly rate: Dropbox (Plus, $9.99 a month), Notion (Plus, $10 per member a month), Calendly (Standard, $10 per seat a month), Zapier (Professional, "Starting from $19.99 /month") and Grammarly (Pro, $12 USD, with no unit or period). Each page has a monthly/yearly toggle; the text shows one number and doesn't tie it to either setting. Calendly marks Standard "Save 17%", which suggests $10 is the annual rate, but the page doesn't say so. Several clear pages also show "{{content}}quot; with no currency code.
Where pages do show both, the annual rate is 17–26% below the monthly one (Zoom, Asana, Figma, Shopify; calculated), and Zapier and ClickUp advertise savings of up to 33%. An answer of "$10 a month" can be right or wrong depending on a toggle the text doesn't record. That is exactly the kind of error that will look like an AI mistake but starts on the vendor's page.
Write both prices next to each other in text, with the period: "$10 per seat a month, billed annually; $12 billed monthly." It costs one line and removes the most common ambiguity.
Sources
3. Promotions hide the list price.
Evidence tier A — HubSpot Starter page and ClickUp pricing page opened. Discount shares are calculated.
HubSpot's Starter page leads with "Starts at $7/month" billed annually and $10 billed monthly. The standard price it shows for both is $20 per seat a month. The discount is, in the page's words, for new customers "for a limited time": $7 is 65% off and $10 is 50% off (calculated). ClickUp shows Core at $9 per user a month billed yearly beside a struck-through "Regular price $12", without saying whether $12 is the monthly rate or a pre-promotion price.
A promotional price is a true fact with an expiry date. Any AI answer that quotes $7 will become wrong the day the offer ends, and any answer that quotes $20 is arguably wrong today. For a competitor, the promotion itself is the signal: HubSpot is discounting its entry plan by up to 65% to win new logos.
If you compete with HubSpot at the low end, expect prospects to anchor on $7 a seat. Know that it's a new-customer, limited-time rate and that the list price is $20, and say so in the deal. If you run promotions yourself, keep the list price in the text so it survives the promotion.
Sources
4. Jira's price went up on 13 October. Its pricing page text doesn't show it.
Evidence tier A — Atlassian pricing-tables page and Jira pricing page opened. Change percentages are calculated.
Atlassian's cloud list-price tables show Jira Standard for 1–100 users rising from $9.05 to $9.70 per user a month, up 7.2% (calculated), and the 10-user annual price from $900 to $950, up 5.6% (calculated), effective 13 October 2026 for new and existing subscriptions. On 31 Oct, the main Jira pricing page still showed no prices in its text, as on 12 Oct. The new price is readable only in the separate list-price table.
The one company in our set that actually changed its entry price between our two readings is also one whose pricing page shows a reader no price at all. An AI that wants to answer "what does Jira cost?" has to find a licensing table most buyers never visit, or fall back on an older figure from somewhere else.
If you sell against Jira, use $9.70 per user a month (1–100 users, monthly) from 13 Oct, and expect buyers and AI answers to quote $9.05 for a while. If you change your own prices, publish the new number on the page buyers actually land on, in text.
Sources
5. In 19 days, 5 of 24 pages read differently.
Evidence tier A — Readings of 12 Oct and 31 Oct compared. Some changes may come from our reader, not the page; we say which.
Between 12 and 31 October, five pages changed status. Figma became clear: it now shows $12 per editor a month billed annually and $15 billed monthly. Calendly and Zapier stopped being clear: their prices ($10 and $19.99) are unchanged, but the text no longer ties them to a billing period. Canva's main pricing page could not be read on 31 Oct because its robots.txt disallows it; its Pro page shows $119.99 paid annually or $12.99 monthly, which we can't yet reconcile with the US$144 a year we read on 12 Oct. Box's page showed no prices in its text on 31 Oct, against a price with gaps on 12 Oct.
Only Jira's change is a confirmed price move. The rest are changes in what a reader can see, and for Box and Canva we can't rule out that our reader, not the page, is what differed. That is the point: a single reading of a pricing page is a snapshot, and the snapshot moves even when the price doesn't.
Don't build a battlecard from one visit to a competitor's pricing page. Keep a dated copy each time you check, and re-check before you quote a number in a deal.
Sources
6. What this means for the AI test.
Evidence tier M — Our own design decision, stated before any AI answer is collected.
The AI test asks ChatGPT, Gemini and Perplexity four questions about these 24 companies: entry price and free plan for all 24, and latest fiscal-year revenue and headcount for the 14 public ones. That is 76 questions per engine. On the pricing question, only the 12 clear pages give a single right answer to grade against. For the other 12 we will publish what the page says, record what each AI says, and label the result "source unclear" instead of right or wrong. Revenue and headcount come from SEC filings, which state their figures plainly, so those 28 questions per engine are fully gradable.
Commercial fact-check tools, such as Peec's and Profound's, grade AI answers against facts the brand types in. Grading against the public page is stricter and fairer to the AI: where the page itself is unclear, an AI's wrong answer is partly the vendor's doing.
When AI gets your price wrong, check your page before you blame the model.
Sources
Every claim links to its source. Spotted something wrong? It will be fixed and logged on the corrections page. Also published on Beehiiv.