Table of Contents
We asked an AI agent (Lindy) to price 20 AI tools. It got 89% right, and every wrong price was the same mistake
Can an AI agent do your pricing research? On October 19, 2026 we gave Lindy, the no-code agent we cover in our Lindy review, the job a lot of people now hand to agents: find the current monthly price of every paid plan for 20 AI tools, from each vendor's own pricing page. Then we opened every one of those pages ourselves and checked all 73 rows it returned.
- 65 of 73 rows were right (89%). 5 were partly right and 3 were wrong. Lindy never skipped a paid plan.
- Every wrong price had one cause. The pricing page opened on annual billing, or in euros, and Lindy reported what it saw as a US monthly price. Make, Notion and Grammarly all tripped it, even though our prompt asked for monthly billing.
- It caught real news. Lindy flagged that Relay.app has shut down and that Windsurf's pricing page now redirects to Devin's.
- One batch cost 15 times more than the next. Five chatbot and coding tools took more than 15 minutes and 241 credits. The other batches took 14 to 66 seconds.
Add our warm-up test on three more tools and the total is 86 rows: 75 right, 5 partly right, 6 wrong. The short version: an agent like Lindy is a fast first draft for pricing research, and a risky final answer for any vendor with a billing toggle.
| Test setup | |
|---|---|
| Date | October 19, 2026 |
| Agent | Lindy Assistant in Lindy chat, mode "Auto", on the Lindy Trial plan |
| Warm-up test | Cursor, Zapier, n8n (13 rows) |
| Main test | 20 tools in 4 batches of 5 (73 rows) |
| Prompt | One per batch, same wording, tool names swapped (prompts) |
| Checked against | Each vendor's pricing page, fetched from a US location, billing toggles clicked |
| Data | results.csv, Lindy's raw answers |
The scorecard
| Tool | Batch | Rows | Right | Partly | Wrong | What went wrong |
|---|---|---|---|---|---|---|
| Cursor | Warm-up | 6 | 6 | 0 | 0 | |
| Zapier | Warm-up | 3 | 3 | 0 | 0 | |
| n8n | Warm-up | 4 | 1 | 0 | 3 | Converted euro prices |
| GitHub Copilot | A | 6 | 6 | 0 | 0 | |
| Claude | A | 6 | 5 | 1 | 0 | Put a 20-seat minimum on Team |
| ChatGPT | A | 9 | 9 | 0 | 0 | |
| Perplexity | A | 5 | 5 | 0 | 0 | |
| Windsurf (now Devin) | A | 5 | 5 | 0 | 0 | |
| Make | B | 4 | 1 | 0 | 3 | Prices match neither billing option |
| Gumloop | B | 2 | 2 | 0 | 0 | |
| Relevance AI | B | 3 | 3 | 0 | 0 | |
| Relay.app | B | 1 | 0 | 1 | 0 | Shutdown right, old Team price wrong |
| Pipedream | B | 2 | 2 | 0 | 0 | |
| Exa | C | 2 | 2 | 0 | 0 | |
| Tavily | C | 3 | 3 | 0 | 0 | |
| Firecrawl | C | 5 | 5 | 0 | 0 | |
| Apify | C | 4 | 4 | 0 | 0 | |
| Browserbase | C | 3 | 3 | 0 | 0 | |
| Notion | D | 3 | 1 | 2 | 0 | Yearly price labeled monthly |
| Grammarly | D | 2 | 1 | 1 | 0 | Annual rate labeled monthly |
| Jasper | D | 2 | 2 | 0 | 0 | |
| Otter.ai | D | 3 | 3 | 0 | 0 | |
| Fireflies.ai | D | 3 | 3 | 0 | 0 |
Batch C, five web data APIs, was perfect. So were the five chatbots and coding tools in batch A, apart from one seat-minimum detail. The trouble sat in two places: automation tools and productivity apps.
Every wrong price was a billing-period mistake
Here are the four tools Lindy got wrong or half wrong on price, next to what the vendor's page says.
| Tool | Lindy said | The page says | What happened |
|---|---|---|---|
| Make Core / Pro / Teams | $9 / $16 / $29 billed annually; $10.59 / $18.82 / $34.12 monthly | $12 / $21 / $38 billed annually; $16 / $28 / $51 monthly | The page opens on annual billing. Lindy's numbers match neither setting |
| Notion Plus / Business | $10 / $20 per member, "monthly billing" | $10 / $20 billed yearly; $12 / $24 billed monthly | The page opens on "Pay yearly". Lindy kept the yearly price and called it monthly |
| Grammarly Pro | $12 a month, "monthly billing" | $12 shown with no period; the page's plan data has $30 monthly, $144 a year | $12 is the annual plan divided by 12. Paying monthly costs 2.5 times as much |
| n8n Starter / Pro / Business | about $22 / $56 / $747, converted from 20 / 50 / 667 euros | $20 / $50 / $800, billed annually, on the US page | Lindy got the euro page and did its own currency math |
Our prompt for the main test said, word for word, "monthly billing if shown, otherwise say annual". Lindy followed that for Jasper, Otter.ai, Fireflies.ai, Pipedream and Claude, where it gave both prices. On Notion and Grammarly it labeled the annual figure as monthly anyway. On Make it went further and reported a "15% discount for annual billing". The page says "Save 25%".
Our Lindy workspace, captured October 19, 2026.
n8n is the same mistake in a different coat. The n8n page Lindy read showed euros, billed annually. It converted at about 1.12 dollars per euro and labeled the result USD. n8n's US prices are not a currency conversion of its euro prices: Business is 667 euros or $800, per n8n's pricing page.
Our Lindy workspace, captured October 19, 2026.
The lesson for anyone using an AI agent for pricing research:
- Ask for both billing periods, each with its label. "Monthly price" alone gets you whatever the page shows first, and most pages show annual first.
- Force USD, and forbid conversion. If the page shows another currency, the agent should say so and stop, not do the math.
- Check every page that opens on annual billing, or shows a price with no period, yourself. Lindy read some of those right (Fireflies.ai, Firecrawl), but every one of its billing mistakes happened on one.
The other slips
Two partly right rows had nothing to do with billing. Lindy put a 20-seat minimum on Claude's Team plan; Claude's pricing page puts that minimum on Enterprise and describes Team as for teams of 2 to 150. And it gave Relay.app's last Team price as $69 when the last archived pricing page we found shows $59.
Lindy also adds notes under its tables, and those were shakier than the tables. Of 16 side claims we checked, 10 were true, 4 false, 1 partly true and 1 unverifiable. The worst: it said Otter.ai's Business plan is custom-quote, two lines after its own table priced it at $30 a month.
What it caught
The notes also held the most useful finds of the day.
- Relay.app is gone. Lindy said Relay.app shut down in September 2026 with all data deleted. relay.app now says exactly that, and the archived shutdown notice gave September 14, 2026 as the date for paying customers.
- Windsurf has no pricing page of its own. windsurf.com/pricing redirects to devin.ai/pricing, and Lindy priced the Devin plans, all five correctly.
- Notion AI is no longer a separate add-on. Lindy noted it is bundled into Notion's plans, which matches Notion's pricing page.
- Perplexity shows annual prices only. Lindy said so, then filled in the monthly prices from Perplexity's help center. All of them checked out.
Lindy also reported that OpenAI's pricing page sat behind a bot wall and used OpenAI's help center instead. Our own fetch of the page loaded fine, so we can't confirm what Lindy's fetcher saw. Its ChatGPT prices were 9 for 9 either way.
Time and credits
| Batch | Tools | Time | Credits | Right |
|---|---|---|---|---|
| Warm-up | Cursor, Zapier, n8n | about 47 s | about 5 | 10 of 13 |
| A | GitHub Copilot, Claude, ChatGPT, Perplexity, Windsurf | more than 15 minutes | 241 | 30 of 31 |
| B | Make, Gumloop, Relevance AI, Relay.app, Pipedream | 66 s | 16 | 8 of 12 |
| C | Exa, Tavily, Firecrawl, Apify, Browserbase | 14 s | 2 or fewer | 17 of 17 |
| D | Notion, Grammarly, Jasper, Otter.ai, Fireflies.ai | 15 s | 2 or fewer | 10 of 13 |
Batch A is the outlier. Lindy's log showed "Worked for 903 seconds" partway through, then two more working stretches. It hunted for pages it could read, chased Perplexity's monthly prices into the help center and confirmed the Windsurf redirect. It got the answers right, and it spent 241 credits doing it.
A credit costs about a cent. Lindy's paid plans start at $29.99 a month for 3,000 credits, and top-ups are $10 per 1,000 credits, per Lindy's pricing page. So batch A cost about $2.41, against about 16 cents for batch B.
Our Lindy workspace, captured October 19, 2026. The batch A task read 241 credits right after the run and 242 by the time we took this screenshot.
To keep a pricing task from running away: send small batches (we would cap it at three to five tools), tell the agent to stop after one pass per vendor and report what it could not read, and check the task's credit count before you queue the next one. Two of Lindy's other tasks that day, a 150-message Slack recall and a weekly routine dry run, used 12 and 9 credits.
How we checked
The prompt. The warm-up asked for "the current monthly prices for every paid plan of Cursor, Zapier and n8n" with a source URL per row. The main test added "in USD" and "monthly billing if shown, otherwise say annual". Every batch used the same wording; only the five tool names changed.
The check. For each tool we fetched the vendor's pricing page through a US-located headless browser, with JavaScript rendered, and clicked the monthly/annual toggle where there was one. Where a page hid prices, we read the vendor's own help center or pricing data. For Relay.app we used the Internet Archive.
The grading. A row is correct when the price and its billing period match the vendor page. Partly means the number appears on the page but is labeled with the wrong period, or a detail in the row is wrong. Wrong means the price is not what a US buyer pays. Every row, with the vendor price and the reason for its grade, is in results.csv.
Run it yourself
- Sign up at lindy.ai. New accounts start with $50 in free credits, per Lindy's pricing page, and our most expensive batch used 241 credits, about $2.41.
- Open a new chat in Lindy, leave the mode on Auto and paste the prompt below with your own five tools.
- Open each vendor's pricing page in a private window, click the monthly/annual toggle, and compare row by row.
This is the prompt we would use next time. It adds the three fixes from our misses; we have not rerun the test with it yet.
For each tool below, open the vendor's own pricing page. For every paid
individual or team plan, report:
- the price billed monthly, per month
- the price billed annually, per month
- the exact billing label shown next to each price
If the page has a monthly/annual toggle, read both states. Report USD only.
If the page shows another currency, write the currency and do not convert.
If a figure is not on the page, write "not shown". Give one table with tool,
plan, monthly-billed price, annual-billed price, label, and source URL.
Tools: Make, Notion, Grammarly, n8n, CursorIn our run most rows came back right. Check every page that opens on annual billing or shows a foreign currency yourself.
What this test does not show
- One agent, one day. We tested Lindy only, on October 19, 2026. Other agents, or Lindy on another day, may slip in different places.
- 23 tools. 20 in the main test plus 3 in the warm-up. Enough to show a pattern, not a rate you can count on.
- Lindy's fetcher, not ours. Lindy reads pages from its own servers, and we can't see where they are. The n8n euro page suggests at least one fetch came from outside the US.
- Pricing pages change. Prices here are as of October 19, 2026. The raw vendor prices we checked against are in the data download.
Download the data
- results.csv: all 86 rows (batch, tool, plan, Lindy's price, vendor price, verdict, reason)
- results.json and side_claims.json: the 73 main-test rows with evidence, and the 16 side claims
- Prompts: warm-up, A, B, C, D
- Lindy's raw answers: warm-up, A, B, C, D
- charts.py: the script behind the charts
For another Pondero Labs test with the same method, see Firecrawl vs Tavily vs Exa, where we ran 50 identical searches through three search APIs.
What changed
- October 19, 2026: First published.