Logins and paywalls: the agent cannot sign up

The wall an agent cannot climb

People handle walls: they log in, they subscribe, they accept the cookie banner and scroll. An agent fetching on a user's behalf does none of that. It sends one request without credentials, gets whatever the server returns to a stranger, and moves on. If that response is a login form, a subscription prompt or an HTTP 401, the page is absent from the answer and the answer goes to a source that was readable.

This is not an argument against paywalls. It is an argument for deciding deliberately which part of each page a stranger can read, because that part is your entire presence in AI answers.

What Qomvia looks for

The Main content is readable without an account check under Machine access fetches a content page as a declared crawler. An HTTP 401 or 402 is a fail. If the page contains login or subscription wording and less than about 1,200 characters of readable text, it also fails, because the response reads as a wall rather than as content. Wall wording alongside a substantial body is a partial: the teaser model, which works.

The check is worth 2 of 100 points, small on purpose. The damage of a hard wall is not the lost points; it is that every legibility check downstream (extractable main content, headings, metadata) is evaluated against a login form.

Patterns that keep the citation

• Metered or teaser access. Serve the headline, standfirst and the first few hundred words to everyone, then gate. The agent gets enough to attribute and quote; the reader still has to subscribe for the rest.
• Structured data for the whole article. Publish Article JSON-LD with headline, datePublished, author and a description, and mark the gated section with isAccessibleForFree: false and hasPart pointing at the paywalled cssSelector. This is the schema.org pattern Google documents for subscription content, and it tells any reader what is public and what is not.
• Public FAQ and summary pages. For SaaS and B2B products, keep pricing, feature descriptions and documentation outside the app. A login-only knowledge base is a common reason a product is never named in answers about its own category.
• Do not gate the homepage. A 'Sign in to continue' interstitial on / fails every check at once.

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Why the cantonal tax model changed",
  "datePublished": "2026-07-14",
  "isAccessibleForFree": false,
  "hasPart": {
    "@type": "WebPageElement",
    "isAccessibleForFree": false,
    "cssSelector": ".paywalled"
  }
}

Gated does not have to mean silent

A site that gates everything can still be discoverable through the machine layer: a public llms.txt describing what the site covers, a feed of headlines, and a stated automated-access policy that tells platforms how to request access. None of that replaces a readable page, but it turns a blank into a known entity.

Questions

Can I show full articles to AI agents and a paywall to people?

Technically yes, by allowlisting declared agents. Most publishers do not, because a user-fetch agent then quotes the full text back to a non-subscriber. The teaser model with structured data is the common compromise.

Does a cookie banner count as a wall?

Only if it replaces the content in the HTML. Most banners are an overlay on top of a fully served page, which agents ignore. A consent gate that withholds the page body until accepted is a wall.

We are a B2B product and everything useful is in the app. What should be public?

Pricing, feature pages, documentation and a changelog. Those are the pages an assistant reads when someone asks it to compare tools in your category.