Zero-Click Is the Default Now. Here's What to Measure and Fix in 2027.

Planning season brings the same question to most website teams: what should we invest in on the site next year? This year the question has a new twist. More people now get their answers from AI summaries and assistants without ever clicking through to a website.

That does not make the website matter less. It changes what the website has to do. Even when nobody clicks, your site is still the source of the facts AI systems repeat about you. So the question for 2027 is whether those systems can find your site, understand it, and cite it accurately.

Where zero-click stands

Zero-click is now the normal outcome of a Google search. SparkToro's 2026 analysis of Similarweb clickstream data puts the US rate at about 68 percent. Pew Research Center found that people clicked a result on 8 percent of searches when an AI summary appeared, compared with 15 percent without one. Inside AI Mode, Seer Interactive found about 93 percent of sessions ended with no click to an outside site.

The losses are not spread evenly. The steepest drops hit simple informational questions that an AI summary can answer in full. Searches for a specific organization, service or product hold up better, because people still have to buy, book or apply somewhere. That makes it more important that AI answers describe you correctly when your name does come up.

What AI systems need from your site

To answer a question about you, an AI system has to work out who you are, what you offer and where, and which of your pages answers the question. It gets those facts from three places: the content itself, the links between your pages, and the structured data in your page code.

Structured data is the part most sites neglect. It is code, usually in a format called JSON-LD, that states facts in a form machines read directly. This is an organization. This is its address. This page answers these questions. This service is offered in these places. Visitors never see it, but search engines and AI systems use it to connect facts without guessing.

Most CMS sites already have some structured data, because SEO plugins add it by default. The trouble is that the defaults say very little.

What an audit turns up

This fall we completed an AI visibility audit for a regional fiber internet provider. We crawled all 332 indexable pages and more than 27,000 internal links. Then we checked the site against the six searches the company most wanted to win. The numbers belong to that site, but these are the patterns any CMS site should check for.

• A thin organization record. The sitewide Organization markup was the plugin default: a name, URL, logo and three social profiles. There was no address, phone number, service area or description, and nothing identifying the company as an internet provider. This record is where machines look first for basic facts about a business.

• Two companies where there should be one. Each of the 176 location pages carried a second, hand-built JSON-LD block. Its provider entity had no @id, so it never connected to the main organization record. To a machine, every location page described a separate company. The most useful local service details sat in that disconnected block.

• Answers without markup. The site had 875 visible question-and-answer blocks across 201 pages, and none used FAQPage or Question markup. The answers were written. They just were not labeled as answers.

• No offer or availability data. No page described plans or prices in structured data. None pointed to the address checker, which is the site's main conversion path. Yet price, speed and address-level availability are the first things people ask about an internet provider.

• Pages nothing linked to. 255 of the 332 pages, including every location page, could not be reached by following links from the homepage. They were listed in the XML sitemap, but no page on the site pointed to them.

• Missing answers to priority questions. Four of the six target searches had no page addressing them. That included fiber versus satellite and fiber versus 5G home internet. Several regional competitors of similar size had dedicated comparison pages.

What the structured data fixes look like

Most of these fixes are template changes, so one update reaches hundreds of pages. For the provider above, the structured data work came down to four steps.

• Complete the organization record. Add a postal address, phone number, a contact point with support hours, the service area and a description. Schema.org has no internet service provider type, so use additionalType to point to the Wikidata entry for one. Add the Google Business Profile to the sameAs list.

• Give every entity one identity. Merge stray JSON-LD blocks into the main graph, or have them reference the organization's @id. Give each service a stable @id too. Then every page describes the same company and the same services.

• Mark up questions where they are answered. Add FAQPage markup to new comparison pages as they launch, then to existing pages with page-specific Q&A. Google has limited FAQ rich results to a small set of sites since 2023. The markup still tells machines plainly which questions a page answers.

• Describe the offer and how to get it. Once pricing is managed from one source, attach Offer and price details to each service. Point availableChannel at the address checker. Use LimitedAvailability rather than InStock, because service depends on the address.

One caution from the same audit. Structured data describes content. It does not replace it. Markup on a page nothing links to, or on an answer that does not exist, does very little. So the plan put the linking fixes and the organization record first. New comparison pages came next, with markup added to each page as it is published.

What to measure in 2027

Once the foundations are in place, traffic needs company on the report. A useful 2027 scorecard tracks five things.

• Presence. For a fixed set of questions your audience asks, how often do AI answers mention or cite you? Check monthly, and log the date and the AI model for each answer.

• Accuracy. When AI answers describe you, are the facts right? Old addresses, retired services and outdated prices spread easily once an AI system picks them up.

• Share against competitors. Appearing on a third of your questions is weak if a competitor appears on most of them.

• Referral quality. AI referrals are still small. Measure what those visitors do once they arrive.

• Demand for your name. Branded search and direct traffic are slow signals. Over a year, they show whether being seen in AI answers builds recognition.

Getting a baseline before targets are set

An audit should end with a baseline and a plan, not a score. The one above produced 19 findings, each rated for effort and impact, and a recommended order that started with low-effort template fixes. That is the shape of the AI visibility audits we run: a full crawl, a structured data review, the target questions you care about, and a ranked list of what to fix first.

Whether you do it with us or in-house, do it before the 2027 targets are set. Search has not stopped mattering. More of its value now happens on the results page, and your site has to be ready to be understood there.