Skip to content
← Back to Blog

A Member Sent Software to Register for Your Conference. It Could Not.

Cloudflare scored 200,000 domains for agent readiness. Fewer than 15 carried the newer agent signals. Nobody is ahead of you on this one, which is the rare good news in an AI post.

The AI question associations have been asking is what happens when a machine answers a question about you instead of sending someone to your website. That is the search question, and it is the one in every trade article.

Here is the question nobody is asking. What happens when a member points their assistant at your site and tells it to handle something? Renew the membership. Pull last year's conference proceedings. Register three people from the office. The assistant arrives at your website and tries to do it.

Where this sits. This is the second of five posts on AI and association websites, covering the surface almost nobody has looked at. The Cloudflare figures below are from its own published research and engineering blog, read directly rather than from coverage of them.

Almost No Website Is Ready for This

Cloudflare launched an Agent Readiness score in April 2026 and scored 200,000 domains from its own Radar data. The results are not close.

  • robots.txt is nearly universal. 78 percent of sites have one, and Cloudflare notes that the vast majority are written for traditional search engine crawlers rather than for AI agents.
  • Almost nobody has said anything about AI in it. Only 4 percent of sites have declared AI usage preferences in robots.txt at all. The file exists on four sites in five and says nothing about AI on ninety-six in a hundred.
  • The newer signals are effectively absent. Cloudflare reports the newer agent standards, MCP server cards and API catalogs, appearing on fewer than 15 sites in the entire 200,000-domain dataset.

Why this is good news. Read that last number again. Fifteen sites out of two hundred thousand. For an association board that has spent three years hearing that the organization is behind on technology, this is the rare area where nobody is ahead, and where a modest amount of work puts you in a category with almost no occupants.

This Is Not the Scraping Argument

Associations have already had one fight about bots, and it was about training. Somebody asked whether anything could be done about AI companies using your content, a switch got flipped, and in more than one case the switch also blocked the crawlers that feed ordinary search.

An agent visiting for a member is not a scraper. An agent acting for a member is a different thing entirely. It is not collecting your content to train on. It is one of your own members, using a different tool to do something they were always allowed to do. Blocking it is not protecting your content. It is refusing a member.

And the plumbing now knows the difference. The infrastructure now distinguishes these. Cloudflare introduced signed agents, using Web Bot Auth and HTTP message signatures so that agents directed by an actual person can prove which platform they come from. Its own framing is that these agents perform user-directed tasks that many sites want to permit rather than reject. The named examples include ChatGPT Agent, Browserbase, Anchor Browser and Goose.

Where Association Sites Break First

We have walked association sites as an agent would. The failures are not exotic and they are the same failures that make the site hard for people, only less forgiving.

  • The login. An agent cannot answer a security question, cannot receive the code sent to a member's phone, and cannot get past a challenge designed to prove humanity. Everything behind the member wall is therefore unreachable, which is most of what a member would send an agent to do.
  • The registration flow. Multi-step registration flows that depend on a session, hidden state and a payment page in a new window fail for the same reasons they fail for a member on a phone with a flaky connection, only immediately.
  • Content that is layout rather than data. A conference schedule rendered as a grid of styled divs with no structured data is a picture of a schedule. An agent asked which sessions cover a topic has nothing to read.
  • Documents instead of pages. Dues tables inside a PDF, member benefits inside a flipbook, deadlines inside an image. These have always been bad for search and worse for members, and an agent simply cannot see them.
  • The blanket bot rule. Aggressive bot rules written during the scraping conversation frequently refuse anything that is not a named search crawler, which now includes your member's assistant.

What Good Looks Like

Nobody needs to rebuild anything for this. The work is mostly making explicit what is already implied:

  • A robots.txt that has an opinion about AI. Say what is permitted, by whom, for what. Four percent of sites have done this. It is an afternoon.
  • Content a machine can parse. Events, courses, job postings, articles and prices marked up so a machine reads them as those things. This is the same structured data work that serves search, which is why it is the highest-return item on the list.
  • A bot policy with two categories, not one. Distinguish an agent acting for a member from a crawler collecting training data, and decide those separately. Treating them as one category is how you end up refusing your own members.
  • The important things as pages. Dues, deadlines, eligibility, benefits and the event schedule as real pages with real text, not as a downloadable file. Every audience benefits and the agent is simply the least forgiving one.
  • Flows that survive being automated. If a member can complete something in a browser, the path should not depend on a mouse hover, a drag, or a step that only exists visually.

The Part Where We Argue With Ourselves

There is an honest objection to all of this, and it deserves stating rather than skipping.

The volume is unproven. Nobody has published reliable figures on how many association members are actually sending agents to association websites today. We have not seen it, and we would not believe a vendor that claimed to have it. It is plausible that for your organization the answer this year is almost none.

Which is why the recommendation still holds. Almost everything on the list above is work you should do anyway. Structured data serves search. Content as pages rather than PDFs serves members and serves accessibility. A bot policy that distinguishes categories fixes the problem where your security settings are quietly blocking Google. A registration flow that survives automation is a registration flow that survives a bad phone connection. There is no version of this list that is wasted if agents turn out to matter less than expected, which is not something we can say about most AI spending.

Ask Your Own Site the Question

Open an assistant you already pay for, point it at your website, and ask it to find your dues structure, your next conference and how someone joins. Then ask it to actually start the join process. Watch where it stops. That takes fifteen minutes and tells you more than any vendor assessment will.

If you want the fuller version, we run association sites against the same four areas Cloudflare scores, discoverability, content, bot access control and capabilities, and send back what an agent can and cannot do on your site today, which of the failures are configuration, which are content, and which are genuinely a build. Most of what we find is the first two. This is part of the ongoing work in our partnership plans rather than a separate AI engagement, because the standards in this area are eighteen months old and will not stay still.

83 Creative

We're a web development studio that works exclusively with trade associations, professional societies, and membership organizations.

← Previous Article Your City’s Website Deadline Depends on One Census Number. Check Which One Is Yours.