Kapiway Research · August 2026

The AI search land grab has already started. We have the receipts.

On August 7, 2026 we ran 50 of the biggest SaaS marketing sites through Kapiway's own audit engine: the same crawl-based checks our free AI Readiness scan runs on any site. We expected to find neglect. We found the opposite: the leaders are quietly positioning for AI search, and the real gap is not where most people think.

31 of 50

ship a real llms.txt on their main domain

4 of 50

restrict AI crawlers in robots.txt

29 of 50

score weak (50 or less) on structured data

37 of 50

have images missing alt text

Finding 1: llms.txt went mainstream while nobody was looking

62% of the sites we scanned (31 of 50) serve a real llms.txt file on their main domain: a plain-text index that tells AI crawlers what the site is and where its important pages live. Not docs subdomains, not auto-generated stubs. Real, curated files on the primary marketing domain, at Stripe, Slack, Salesforce, Shopify, HubSpot, Notion, Webflow, Zapier and 23 more. Slack's file literally opens with "Welcome, humans and bots alike."

For calibration: when we checked 20 well known SEO tools four days earlier, only 35% had one on their main domain. The SaaS leaders have out-adopted the SEO industry itself.

Finding 2: almost nobody locks AI out, but four sites are curating it

Only 4 of the 48 sites whose robots.txt we could read restrict any AI crawler (two sites, gusto.com and mercury.com, refuse automated fetches entirely). Loom blocks GPTBot outright. Calendly blocks CCBot everywhere except two pages it clearly wants read: /llms.txt and /pricing. Canva curates five training crawlers with allow exceptions and fully blocks Bytespider. And Figma has written the most deliberate AI policy in the study: the training crawlers (GPTBot, ClaudeBot, Google-Extended, CCBot) are blocked entirely, while the answer engines' live agents (OAI-SearchBot, CHATGPT-User, PerplexityBot, Claude-SearchBot) are granted access through roughly 8,700 allow rules covering hand-picked pages: home, pricing, release notes and localized versions. That is not opting out of AI search. That is curating exactly what AI may read: feed the answers, starve the training sets. The remaining 44 readable sites simply leave the door open.

Finding 3: the real moat is structured data, and most sites are weak there

Here is the gap nobody talks about. 58% of these sites (29 of 50) score 50 or below on structured data: the schema markup that helps machines actually understand what a page is, who it is for, and what it offers. The file that says "AI, please read me" is everywhere. The markup that makes the reading useful is not. Add the hygiene gaps (37 of 50 have images missing alt text, 45 of 50 have title tag length problems) and the pattern is clear: adoption of the visible signal is ahead of the fundamentals.

Every site, every number

Scores are from Kapiway's crawl-based audit (up to 6 pages per site, public pages only, Aug 7 2026). llms.txt = real text file served at /llms.txt on the main domain. Blocked = AI crawlers disallowed sitewide in robots.txt. Rerun any site through the free scan to check the current state.

Site	llms.txt	Blocks AI bots	On-page	Structured data	Technical
15five.com	no	no	75	100	100
activecampaign.com	no	no	75	50	100
airtable.com	no	no	75	50	94
asana.com	yes	no	64	50	89
basecamp.com	no	no	62	100	100
box.com	no	no	62	50	100
brex.com	yes	no	100	100	100
buffer.com	no	no	73	100	100
calendly.com	yes	CCBot	79	58	100
canva.com	no	GPTBot, ClaudeBot, CCBot, anthropic-ai, Applebot-Extended, Bytespider	50	50	33
clickup.com	yes	no	62	58	100
deel.com	yes	no	62	50	67
docusign.com	no	no	62	50	100
dropbox.com	yes	no	75	100	100
figma.com	no	GPTBot, ClaudeBot, Google-Extended, CCBot	75	100	100
framer.com	yes	no	100	100	100
freshworks.com	yes	no	50	50	33
ghost.org	no	no	79	92	94
grammarly.com	no	no	88	50	100
gusto.com	no	no	50	50	33
hootsuite.com	yes	no	88	50	100
hubspot.com	yes	no	88	50	100
intercom.com	yes	no	100	50	100
kit.com	no	no	75	50	100
klaviyo.com	yes	no	75	50	100
lattice.com	yes	no	75	50	100
linear.app	yes	no	71	50	100
loom.com	yes	GPTBot	88	100	100
mailchimp.com	yes	no	83	100	100
make.com	yes	no	50	50	33
mercury.com	no	no	50	50	33
miro.com	no	no	71	75	89
monday.com	yes	no	42	50	72
notion.so	yes	no	88	50	100
pipedrive.com	no	no	50	50	33
ramp.com	no	no	86	100	94
rippling.com	yes	no	88	100	100
salesforce.com	yes	no	75	100	100
shopify.com	yes	no	75	100	100
slack.com	yes	no	75	50	100
sproutsocial.com	yes	no	87	100	100
squarespace.com	yes	no	75	100	100
stripe.com	yes	no	52	50	100
surveymonkey.com	yes	no	100	100	100
typeform.com	no	no	75	100	100
webflow.com	yes	no	75	50	89
wix.com	yes	no	88	100	100
zapier.com	yes	no	71	100	100
zendesk.com	yes	no	88	50	100
zoom.us	no	no	77	58	100

Method, honestly

Sites: 50 widely known B2B SaaS marketing domains, chosen before scanning, none excluded afterward.
Engine: Kapiway's production crawl audit (sitemap discovery, up to 6 public pages per site), plus direct checks of /llms.txt and /robots.txt. No logins, no private data.
llms.txt counts only if the main domain serves a real text file, not an HTML error page. We spot-verified positives by hand.
Scores are category averages of pass rates on standard checks (titles, meta, headings, canonical, schema presence, alt text, HTTPS, status codes and more).
Robots analysis covers 48 of 50 sites; two (gusto.com, mercury.com) block automated fetches of robots.txt itself.
Point-in-time snapshot: sites change. Run any of them through the free scan for today's answer.

Where does your site stand?

The same engine that produced these numbers grades any site free, in about 2 minutes.

Get your free score