On this page
- What agent-ready means
- Why agents read SaaS sites now
- How we tested
- What the agent found
- Can it find your price?
- Can it confirm security?
- Does it know your integrations?
- Can it compare you?
- Can it take the next step?
- The six sites that passed
- Page-by-page checklist
- Test your site in 5 minutes
- WebMCP, llms.txt or an API?
- FAQ
- Sources
More and more often, the first visitor to read a SaaS pricing page is not a person. Buyers ask ChatGPT, Claude or Perplexity to shortlist vendors, and a growing share of them hand the research to an agent that opens each vendor's website, reads it as text and reports back with a table. In October we checked whether AI understands what freshly funded SaaS startups sell, and for 39 of 42 homepages it did. So in November we gave an AI agent the next job on the same companies: build a buyer's shortlist entry using only the website.
I'm Ignat, founder of DIXI. We design and build websites for B2B SaaS, AI and fintech teams from Seed to Series B, so the question I care about is a practical one: can an agent shortlist your company without a sales call? Below are the numbers from our test, what the best sites do differently, and a checklist for each page an agent reads.
- An agent-ready SaaS website states its price, its security status, its integrations and its next step in plain text, on pages linked from the homepage, navigation or footer.
- In our test of 42 B2B software startups that raised in September and October 2026, 21 said nothing about price, 15 gave nothing quotable about security, and only 6 showed a price, a security attestation and their integrations together.
- Most fixes are content and links, not a redesign: a "from" price, one honest sentence about security, integrations named in text, comparison pages in the footer, and a demo page that says what happens next.
- An independent test of 100 B2B products found the same weak spot: agents answered pricing questions from the vendor's own site 79% of the time, against 93% for integrations and 92% for security.
What is an agent-ready SaaS website?
An agent-ready SaaS website is one where an AI agent working for a buyer can find the facts that go into a vendor shortlist, including price, security status, integrations, alternatives and the next step, as plain text on pages it can reach by following links. It doesn't need special files or new code to start with. It needs the facts to be written down where software can read them.
This is a different job from getting mentioned in AI answers. Getting an assistant to name you is what generative engine optimization covers, and we wrote a separate GEO playbook for it. This article is about the step after that, when the agent comes to your site to fill in the buyer's comparison table and either finds the answer or writes "not stated".
| What a buyer's agent looks for | Where it usually looks | What it reports when the fact is missing |
|---|---|---|
| What you sell and to whom | Homepage first screen | A guess from the category |
| Price or pricing model | Pricing page linked in the navigation | "Price not stated on the site", or a number from a directory |
| Security and compliance | Security or trust page, often in the footer | "Not stated", or a claim it can't verify |
| Integrations | Integrations page or homepage text | No answer for the buyer's stack |
| Alternatives | Comparison or "vs" pages | Someone else's comparison |
| Next step | Demo, contact or sign-up page | "Book a demo", with no details |
Why are AI agents reading SaaS websites now?
Because buyers already do their research in AI tools, and those tools have started acting on their own. Cloudflare says more than half of the traffic reaching sites on its network is now automated, and daily requests from AI agents grew by more than 1,700% in a year. At the same time, most B2B software deals are decided before the first sales call.
| Number | What it means | Source |
|---|---|---|
| 1,700%+ | Growth in daily AI agent requests on Cloudflare's network in one year | Cloudflare, Sep 2026 |
| Up to 40% | Drop in human traffic in heavily crawled categories such as computer software and IT services, in less than a year | Cloudflare, Sep 2026 |
| 51% | Software buyers who start research in an AI chatbot more often than in Google | G2, Apr 2026 |
| 69% | Buyers who chose a different vendor than planned because of a chatbot's guidance | G2, Apr 2026 |
| 95% | Deals won by a vendor that was already on the buyer's shortlist on day one | 6sense, 2025 |
The agents themselves got more capable this autumn. OpenAI launched dots at DevDay on 29 September, persistent agents with their own cloud computer that take on ongoing tasks. Chrome has been testing WebMCP, a way for a site to declare actions an agent can call, since Chrome 149, and ChatGPT's desktop browser started using it in August.
One detail from Forrester explains why this matters for a demo-led site. In an August 2026 post, its analysts wrote that agents don't submit "Contact Us" forms. When the facts aren't on the page, an agent falls back to third-party sources or points its human to someone else.
How did we test the startup websites?
We used the same 42 B2B software companies as in our October homepage study, all of which announced a pre-seed to Series B round between 1 September and 27 October 2026. On 2 November an AI agent read each site as text, starting from the homepage, and wrote a buyer's shortlist entry from what it found.
- Sample: 12 companies in AI, dev tools and security, 10 in fintech, legal and insurance, 10 vertical SaaS for physical industries, and 10 in GTM, HR and support.
- Reader: an agent with a web fetch tool that doesn't run JavaScript, which is how AI assistants usually read pages. Vercel and MERJ found that none of the major AI crawlers render JavaScript. It followed links on the site, with at most 8 page loads per company.
- Job: five facts for the buyer: what it is and for whom, the price, the vendor's own security attestation, named integrations, and how to start.
- Grading: only facts in the page text counted. A security product that helps customers pass SOC 2 doesn't count as the vendor having SOC 2, and customer logos don't count as integrations.
- Limits: 42 young companies are not the whole market. Agentic browsers that render pages can see more than a text reader. Our fetch tool refused some direct requests, so some pages were reached through a site search, a few sites were judged on two or three pages, and robots.txt and llms.txt were not checked.
The anonymized results are free to reuse under CC BY 4.0: download the dataset (CSV).
What did the AI agent find on 42 SaaS websites?
Integrations and security were usually there, while prices and comparisons usually weren't. Only 6 of the 42 sites, about 1 in 7, gave the agent a price with a number, a named security attestation and at least three integrations at the same time.
The groups were very different, and not in the way I expected:
| Group | Price with a number | Named security attestation | 3+ integrations | All three |
|---|---|---|---|---|
| AI, dev tools, security (12) | 4 | 10 | 10 | 3 |
| Fintech, legal, insurance (10) | 2 | 9 | 6 | 1 |
| GTM, HR, support (10) | 5 | 6 | 7 | 2 |
| Vertical SaaS (10) | 3 | 2 | 6 | 0 |
The vertical SaaS group had the strongest customer proof in October, with 9 of 10 showing a customer result with a number, and the weakest answers here. My guess is that companies selling to utilities, contractors or hospitals still assume a long, relationship-led sale. Their buyers are getting agents too, though, and an agent doesn't know your sales team.
Can an AI agent find your SaaS pricing?
On two sites out of three, it couldn't find a number. 21 of 42 had no pricing link and no pricing information at all, 6 described plans or a model without a number, and 1 had a pricing page that only said to book a demo. 14 showed a price an agent could put in a comparison table.
That matches a larger test. Kevin Indig and Siteline ran three buyer tasks across 100 B2B products and found pricing was where agents most often left the vendor's site: 79% of pricing answers came from the vendor itself, and pricing produced 77% of all third-party citations. When no real price was disclosed, 45% of runs cited a third party, usually an editorial page or a directory. So if your site doesn't state a price, the agent still reports one, it just takes it from someone else.
I understand why early-stage teams hide prices, because deals are custom and the number changes every quarter. But compare what lands in the buyer's table. Siena says $750 a month for the platform plus $0.90 per automated ticket, and the vendor next to it says "price not stated on the site". We follow our own advice here, and our prices are on our pricing page.
Three sites in the sample had a different problem: conflicting or placeholder prices, such as the same plan at different prices in different parts of the page, or a pricing template that never got its numbers. An agent will quote whichever version it reads first. The sites that worked kept it simple:
- Town lists its plans from free to $199 per user per month on one page.
- Daytona puts usage rates, such as $0.0504 per vCPU hour, right on the homepage.
- Outmarket AI shows $83 per seat per month and says plainly that there's no free trial but a 7-day money-back guarantee.
If you can't publish a full price list, publish what you can stand behind: a "from" price, the pricing unit (per seat, per cluster, per ticket) and what moves the number.
Can an AI agent confirm your security status?
Usually, but it's also where agents can get things most wrong. 27 of 42 named an attestation such as SOC 2 Type II or ISO 27001. The other 15 gave the agent nothing it could quote, and some used wording that sounds like a certification without being one.
Even among the 27, the proof often sat in the wrong place. For at least 7 of them it lived only in a trust portal or on a page about one product, while the homepage said nothing or only something general. Two trust portals came back as empty pages to a reader without JavaScript, because the portal draws everything with scripts.
Among the 15, some had no security information at all and some only mentioned encryption or GDPR. Others used wording that a summary can easily turn into a claim: a "ready" next to SOC 2, a "Certifications" heading where every item was still pending, or badges for ISO standards that are guidance documents you can't be certified against. G2 found that 64% of buyers often run into inaccurate chatbot recommendations, and "pending" next to a SOC 2 logo is exactly the kind of detail that gets lost in one.
The good versions were boring in the right way. Town says SOC 2 Type 2 on its homepage and has a trust page that explains how to get the report, and Daytona's trust center shows a SOC 2 Type II report dated July 2026. One plain sentence with the attestation, the type and the date, on a page linked from the footer, is what an agent can quote.
Does it know which tools you integrate with?
Mostly yes. 29 of 42 named at least three tools they work with. The 13 misses were a mix of sites that named one or two tools or only said something like "connects to your CRM", and logo walls where the logos turned out to be customers or investors.
"Does it work with our stack" is one of the first filters a buyer's agent runs. A row of unlabeled logos gives it nothing to report, while a sentence like "two-way sync with Salesforce and HubSpot" answers the question directly. If you have an integrations page, link it from the homepage or footer and write the names as text next to the logos.
Can it compare you with the alternatives?
Rarely from your own site. G2 found that comparing vendor strengths and weaknesses is the number one way buyers use chatbots for software research (41%), yet only 5 of 42 startups linked a comparison page from their navigation or footer.
The ones that did made it easy. DataAgent has a "Competitive comparisons" block in the footer with pages against Datadog, Elastic and Grafana Cloud, and Restate links "Restate vs Temporal" from its homepage.
What surprised me more is that several companies had already written these pages and then never linked them. At least 6 sites had a page an agent needs, such as a pricing page, "X vs Y" pages, a trust center or an integrations list, that exists but isn't linked from anywhere a visitor would look. A search engine can find those pages, but an agent walking the site from the homepage usually can't.
Can the agent take the next step for the buyer?
Often not. 26 of 42 offered only "Book a demo", and on 21 of the 36 demo or contact pages the agent could read, there was no call length, no specific agenda and no reply time. On at least 7 sites a key button had no destination in the page text at all.
The broken buttons are the easiest fix in this whole study. "Book a demo" pointed to "#", or only worked through a script, so a reader without JavaScript saw a label with nowhere to go. A person with a normal browser never notices, which is probably why it survives. The human side of the same problem, what buyers check before they book, is in why SaaS sites get traffic but no demos.
Two companies showed what good looks like. Punch, a construction software startup, has a page written for AI agents. It points the agent to a plain-text summary, documents a small API where an agent can request a demo or an account for a contractor, and says a human confirms every request by email within one business day. UpSmith's demo page does the same for people: it says the call is 25 to 30 minutes on video, that they'll run a live analysis on your own data and cover pricing and the setup timeline, and that the invite goes out immediately.
Which SaaS websites gave the agent everything?
Six sites showed a price with a number, a named security attestation and at least three integrations. None of them did anything unusual, which I find encouraging: the facts were written as text on ordinary pages that the homepage, navigation or footer linked to.
| Company | What the agent could quote on 2 Nov 2026 |
|---|---|
| Daytona | Usage rates on the homepage, a dated SOC 2 Type II report in its trust center, Docker, Git and VS Code plus SDKs |
| Town | Plans from free to $199 per user per month, SOC 2 Type 2, Gmail, Slack, HubSpot, Salesforce and Notion |
| Namespace | Team $100 and Business $250 per month, SOC 2 Type 2, GitHub Actions, Bazel and Docker |
| Outmarket AI | $83 per seat per month, SOC 2 Type II and ISO 27001, Applied Epic, AMS360 and HawkSoft |
| Lightfield | Starter $89 per user per month, SOC 2 Type II, two-way sync with Salesforce and HubSpot |
| Siena | $750 a month plus $0.90 per automated ticket, SOC 2 Type II, Shopify, Zendesk and Yotpo |
Prices and plans change, so treat this as a snapshot of what their sites said on the day we read them.
What should each page say so an agent can shortlist you?
Go page by page and check that every fact a buyer's agent needs is written as text and linked from somewhere it will look. Every unticked line below is a cell the agent will leave empty or fill from someone else's site.
- A number, a "from" price or a typical range, in HTML text rather than an image or a calculator only
- The pricing unit and what moves the price (seats, usage, modules)
- One canonical pricing URL, linked in the navigation
- The same numbers everywhere: plan cards, comparison table and FAQ
- Each attestation named with its type and date, such as SOC 2 Type II, audited in 2026
- Anything in progress marked clearly as in progress, away from the badges
- A link to the trust portal plus a text summary that works without JavaScript
- Linked from the footer of every page
- The main integrations named in text on the homepage
- An integrations page with names, not only logos
- How each one works: native, API, or through a connector
- "X vs Y" or alternatives pages for the vendors buyers compare you with
- Factual tables with a date, linked from the footer
- Every main button is a real link, not "#" or a script-only action
- The demo page says how long the call is, what it covers and when you reply
- A second path for agents, such as a calendar link or an email address
- A self-serve trial or sign-up, if you offer one, linked from the homepage
- Key text is in the HTML and readable with JavaScript turned off
- robots.txt and your CDN settings allow the AI assistants your buyers use (step 1 of our GEO playbook)
- Buttons and form fields have labels, which Lighthouse's Agentic Browsing audit checks
How can you test your own site in five minutes?
Ask an AI assistant to do what a buyer's agent does, using only your website, and read the answer as the buyer would. Every "not stated on the site" is a gap to fix. Then run two free scanners for the technical side.
You are a procurement assistant building a vendor shortlist for a buyer. Using only [your homepage URL] and pages linked from it, write a shortlist entry: 1. What the product is and who it is for 2. Price (exact numbers if published) 3. The vendor's own security attestations (SOC 2, ISO 27001 etc.) 4. Named integrations 5. How to start and what happens after a demo request Write "not stated on the site" for anything you cannot find. List the URLs you used for each answer.
- Is It Agent Ready? is a free Cloudflare scanner that checks robots.txt, sitemaps and the newer agent standards in under a minute.
- Lighthouse in Chrome DevTools now has an Agentic Browsing category. Much of it is an accessibility check, because Google says agents rely on the accessibility tree as their main view of a page.
Do you need WebMCP, llms.txt or an API for agents?
Not first. These help agents act on a site, but only after the basic facts are on the page. A site with no price in its text won't get better because it adds a new protocol.
WebMCP lets a page declare actions such as "request a demo" as tools that an agent can call instead of guessing at buttons. Chrome is testing it in an origin trial from Chrome 149, and ChatGPT's desktop browser has used it since 25 August 2026. It's worth a look for your demo and sign-up forms once the content is right. llms.txt is cheap to add but optional: our GEO playbook explains why we treat it as hygiene rather than a lever. An agent API like Punch's makes sense if agents already ask to buy from you.
FAQ
What is an agent-ready website?
It is a website where an AI agent working for a buyer can find the facts that go into a vendor shortlist, such as price, security status, integrations, alternatives and the next step, as plain text on pages linked from the homepage, navigation or footer. Special files and protocols help later, but the facts come first.
Do AI agents fill in demo request forms?
Usually not on their own. Forrester analysts wrote in August 2026 that agents don't submit "Contact Us" forms; they report back to the buyer or look for the information elsewhere. That's why the demo page should say what happens on the call and offer a direct path, such as a calendar link.
Should a B2B SaaS show pricing if every deal is custom?
Show what you can stand behind: a "from" price, a typical range or the pricing unit and what moves it. In a test of 100 B2B products, agents fell back to third-party sources in 45% of runs where no real price was disclosed, so a hidden price usually gets filled in by someone else.
Is it OK to show "SOC 2 in progress" on a website?
Yes, if it is true and clearly separated from attestations you already hold. Write the status in a sentence, with the expected date if you have one, and keep it away from certification badges, so a summary can't turn "in progress" into "certified".
Do I need llms.txt or WebMCP to be agent-ready?
No. Neither replaces facts in your page text. llms.txt is an optional summary file, and WebMCP is an experimental way to expose actions such as booking a demo. Both are worth adding after the pricing, security, integrations and demo pages answer a buyer's questions.
How can I see what an AI agent finds on my site?
Give an AI assistant the prompt above with your URL and ask it to use only your site and to list the pages it used. Then scan the site with Cloudflare's Is It Agent Ready tool and run Lighthouse's Agentic Browsing audit for labels and layout issues.
How is agent readiness different from GEO?
GEO is about getting AI assistants to mention and cite you when a buyer asks a question. Agent readiness is about what happens next, when an agent comes to your site to fill in price, security, integrations and the next step. You need both, and our GEO playbook covers the first part.
Sources
- AI agent shortlist test 2026, anonymized dataset · DIXI, Nov 2026
- SaaS homepage best practices 2026: data from 42 startups · DIXI, Oct 2026
- The Internet has a second audience · Cloudflare, Sep 2026
- The Answer Economy · G2, Apr 2026
- 2025 B2B Buyer Experience Report · 6sense
- AI agents can't read your pricing, and it's becoming a revenue problem · Forrester, Aug 2026
- Where AI agents get stuck on your site · Kevin Indig and Siteline, Jul 2026
- DevDay 2026 announcements · OpenAI, Sep 2026
- WebMCP origin trial · Chrome for Developers, Jun 2026
- ChatGPT adds WebMCP site tools · Search Engine Journal, Aug 2026
- Lighthouse agentic browsing scoring · Chrome for Developers
- The rise of the AI crawler · Vercel and MERJ, Dec 2024
- Is It Agent Ready? · Cloudflare
- Company pages quoted: Siena pricing, Town pricing, Punch for AI agents, Restate vs Temporal · read 2 Nov 2026
