A growing share of buying happens before anyone loads your marketing site. The person still pays. Their agent already compared you in documentation, tried a call, and picked a vendor.
The homepage can be sharp and still miss the meeting. The decision moved.
The page you keep polishing
Most teams still treat the homepage as first touch. Hero, proof, pricing teaser, a button that starts a trial. That model assumed the buyer would arrive as a visitor, look around, and then try the product.
Agents do not arrive as visitors. A founder asks an assistant which email API to use. A developer tells a coding agent to add webhooks. The software goes to reference pages, OpenAPI files, GitHub READMEs, and example curls. Your campaign page never gets a chance to be convincing.
This is not "zero-click" as a slogan. It is a shorter path: task in, vendor out. If your public reference is enough to try the job, the homepage is optional. If it is not enough, the agent does not escalate to your hero section. It tries the next vendor.
I used to start Monday with homepage conversion. It felt like the honest number. Then I started asking what last week's 401s were attached to. A lot of them had no session behind them at all.
The email API that lost in a chat
A founding engineer asked Claude for a transactional email API he could wire up before lunch. Two vendors came back. He never opened either homepage.
Vendor A had a beautiful marketing site. The agent opened the docs, hit "create an API key," and found a dashboard login. No test key in the quickstart. The next step was a browser.
Vendor B put a sandbox key in a code block and showed the exact POST /v1/send body. The agent ran the curl, got a 200, and pasted the snippet into the repo. The engineer shipped Vendor B that afternoon.
Vendor A's analytics that day were fine. Homepage bounce looked normal. Signup was a little light. Nobody on that team would have called it a lost deal. There was no deal record to lose. The comparison finished in a sidebar. Marketing never got a session to attribute.
That is the homepage lie. You can win the page and lose the job.
Where the comparison actually happens
The buyer still has taste, budget, and a brand they will defend in a meeting. They also have less patience for walking an assistant through your funnel. So the assistant uses whatever you left in public.
Usually that is not the narrative on /. It is the first successful request. A working example beats a paragraph about "developer-first." A named error beats a claim about reliability. A token the agent can obtain without a human inbox beats a story about security.
None of this makes your homepage useless. Humans still need a place to understand the company. Investors, candidates, and late-stage buyers still land there. The mistake is treating that page as the start of every serious evaluation.
Marketing can keep winning the narrative and still lose the first successful request. That request is the shortlist now. If you only instrument the story, you will keep celebrating traffic that arrived after the choice was already made.
Software already has a new class of user. That user does not owe you a visit.
What you watch | What the agent did |
|---|---|
Homepage session | Never loaded |
Trial signup | Used a sandbox key from the docs |
Bot traffic on | Compared two vendors in one chat |
401 on token create | Could not mint a key without a browser |
"No conversion" | Converted to a competitor via a snippet |
If your dashboard only tells the left column, you will keep funding the page and starving the path.
Your funnel can look healthy while the decision leaves
Pageviews and trial starts still move when humans show up. That is why this is easy to ignore. The human funnel is not fake. It is incomplete.
Agent evaluation shows up as fragments: a docs fetch with no referrer you like, a burst of 401s, a single successful call from a cloud IP, then nothing. Security may shrug. Growth may never see it. Product may call it crawler noise.
The commercial fact is simpler. Someone asked software to pick a tool. Software tried you. You were not choosable in the time it had. That attempt was demand. It just never became a session you would brag about in standup.
You do not need a new religion around measurement to accept that. You do need a place that treats those fragments as one attempt. Analytics built for people in browsers will not do that work for you. It will keep scoring the pages the agent skipped.
Do not wait for a clean "agent channel" report. Tomorrow is enough.
What you can check tomorrow
Pull last week's 401s on token, OAuth, and key-create routes. Ignore the ones that are clearly your own CI. Look at the rest.
Ask three questions. Did the request follow a docs or OpenAPI fetch? Did it retry with a different header and then stop? Did a customer who later appeared in CRM already know your endpoint names on the first human call? You will not get courtroom proof. You will get a short list of attempts that never became homepage sessions.
Then open your own quickstart the way a literal intern would. If the first credential requires a human inbox, a captcha, or a dashboard click, you have a page that marketing cannot save. If the first error is a shrug, you have the same problem with a 200 on the docs page.
Export the week. Sort by path. Read ten rows that are not yours. That is the check.
I still want a homepage that does not embarrass us. I just stopped using it as proof that demand arrived.
Frequently asked questions
Does this mean I should stop investing in the homepage?
No. The homepage still carries brand, hiring, pricing context, and buyers who want a company story before they trust an API. Cut the fantasy that it is the first room every serious customer enters. Fund the reference path with the same seriousness you give the hero.
If they never visit, how do I know we were evaluated?
You look for the leftover: docs hits without sessions, key-create failures, short bursts of "wrong but close" API calls, and customers who arrive later already knowing your endpoint names. Absence of a homepage visit is not absence of a comparison. It is a comparison you were not invited to narrate.
The next customer may already have a vendor. Check the 401s before you rewrite the headline.