Picture someone asking an AI assistant for the best robot vacuum under $400. They get three picks, a one-line reason for each, and a small citation to a review site. That afternoon, they buy the second pick. The review site did the testing and shaped the decision. In its analytics, that person doesn’t exist.
That person is the hardest part of AI search to explain to a dashboard. Most of how we measure SEO assumes value shows up as a click, so a channel where clicks rarely happen looks like it’s shrinking. I’ve come to see it as a brand channel we haven’t learned to measure yet.
That holds for nearly any kind of site: a local service business, an online store, a software company or a publisher. The frame I keep coming back to has three parts: be seen, be believed, be chosen. Below is why the shift is happening, why trust now decides who wins, and what each stage asks of a site.
The answer now lives in the summary
The numbers are stark. In February 2025, Bain & Company reported that about 80% of search users rely on AI summaries at least 40% of the time, and that about 60% of searches now end without the user going on to another site. Bain estimates this has cut organic web traffic by 15% to 25%.
Even skeptics behave this way. In the same Bain research, roughly half of people skeptical of generative AI said most of their queries get answered on the search page without a click. People also use AI assistants for a wide range of tasks: 68% of LLM users use them for research and summarizing information, 48% for news and weather, and 42% for shopping recommendations.
That last number matters for any business that sells something. Product research, comparisons and “best” lists are moving into the answer itself. A product page, a contact form and a checkout all depend on a person arriving, and more often that person stops at the summary. Burt Herman, co-founder of Hacks/Hackers, calls where this ends up “People Zero”: people get answers without ever reaching the site that supplied them.
So the goal shifts. An appearance in an AI answer that never produces a visit can still shape what someone believes and what they buy. The questions are how to win that influence and how to tell whether you’re winning it. I don’t think anyone has a clean answer to the second part yet.
People use AI answers, and they don’t fully trust them
Adoption and trust are moving in opposite directions. In a Gartner survey of 377 U.S. consumers in June and July 2025, 53% said they distrust or lack confidence in AI-powered search results, and 41% said AI overviews make searching more frustrating.
A 2025 Exploding Topics survey of more than 1,000 people found something similar: around 82% said they’re at least somewhat skeptical of Google’s AI Overviews. Yet fewer than 8% said they always follow the links to check the sources.
The doubt reaches past AI. In Gallup and Knight Foundation research, half of Americans said national news organizations intend to mislead, misinform or persuade them.
The result is a two-stage journey:
- Early research People happily take the summary when they want quick background or a definition. The stakes are low and convenience wins.
- High-stakes decisions When money is on the line, people want to check the work. Gartner found that the perceived usefulness of AI summaries drops as people move to evaluating options, and drops further when they’re ready to buy. At that point they look for the original source, the evidence and a business they can trust.
The second stage is where a trusted site still has the advantage. It’s also where most of the commercial value sits.
What makes a source credible
Research on trust offers a useful lens. One of the most widely cited models, from Mayer, Davis and Schoorman (1995), breaks trust into three parts: ability, integrity and benevolence. The same three parts map well onto what people look for in a source online. I call the first one expertise. Dana Mahr and colleagues build on the same model in a 2025 study of public trust in science journalists in the Journal of Science Communication.
| Dimension | What it means | What it looks like on a site |
|---|---|---|
| Expertise | The source knows what it’s talking about | Named experts with real credentials, clear about pages, consistent business details |
| Integrity | The source is honest about how it knows | Published methods and pricing, links to evidence, real customer reviews, sponsored content clearly labeled |
| Benevolence | The source acts in the person’s interest | Useful guides and tools that help whether or not someone buys |
Anyone who has worked on E-E-A-T will recognize this list. Google’s guidance on helpful content asks similar questions about experience, expertise, authoritativeness and trust. The stakes are higher now, because a site has to earn trust in two places: in the answer and on the page someone lands on.
Be seen: make your content easy to find and understand
AI assistants answer questions by drawing on pages from across the web. The first job is to make your pages easy for those systems to read and to attribute to you.
Make pages legible.
- Decide which AI crawlers can access your content, and set robots.txt to match. This is a business choice as much as a technical one.
- Use clean, semantic HTML (
<article>,<section>,<header>), and keep core text out of content that only loads with JavaScript. - Add structured data that states what the page is about, with properties like
@type,about,authorandpublisher. Use the most specific type that fits, such as LocalBusiness, Product, Service or Article. - Put a direct answer in the first sentence or two under each heading, so a passage still makes sense when it’s quoted on its own.
- Write toward the full questions your customers ask, in their words.
Keep your entity data consistent. Harry Clarkson-Bennett’s entity SEO framework is a good model. In short:
- Own Audit every property you control, from your About page and footer to your social profiles and business listings. Make names, descriptions and key details match everywhere. Add Organization or LocalBusiness schema with
sameAslinks to verified profiles. - Describe Give each page one clear subject and headings that answer specific questions. Use Product, Service, Article and Person schema, and link detailed pages to the topic pages above them.
- Prove Show who your people are and why they’re credible: bios, credentials, case studies and the methods behind your claims.
- Earn Build accurate mentions on respected third-party sites. Clarkson-Bennett argues that mentions are better evidence of reputation than links.
- Monitor Track how your brand is described across the web and inside the major AI assistants, and fix inconsistencies at the source.
The cost of skipping this is real. Search Engine Land reported on what it called Google’s “great clarity cleanup,” a 2025 update that removed more than three billion entities from Google’s Knowledge Graph. Entities that are poorly described or weakly corroborated are the most exposed.
Be believed: show your receipts
Visibility helps only if people trust what they see. When someone moves from the summary to a real decision, the page they land on has to earn that trust quickly.
- Show the evidence Link to the data, tests, documents or reviews behind a claim, right where you make it.
- Explain your method Say who did the work, how a product was tested or a result was measured, what the limits are, and where any conflicts of interest sit.
- Be specific about outcomes Case studies with real numbers and time frames, transparent pricing and clear policies answer the questions a careful buyer asks.
- Put the answer first Lead with the essentials, then add depth below for people who want it. The same structure helps a reader with one minute and a system quoting one passage. Newsletter strategist Dan Oshinsky’s 1/5/10 strategy builds a piece to work for a reader with one, five or ten minutes.
Some publishers have gone further and built assistants that answer only from their own reporting. Aos Fatos’s FátimaGPT is a fact-checking chatbot on WhatsApp, Telegram and the web that answers with citations from its archive.
The common thread is verification. A skeptical person trusts what they can check for themselves.
Be chosen: give people a reason to come to you
An AI assistant can summarize a page in seconds. Tools, live data, accounts and communities are much harder to reproduce. Those are the strongest reasons for a person, or an AI agent acting for them, to come to you directly.
- Interactive tools Calculators, quote estimators, configurators and comparison tools built on data only you have.
- Personalization Saved preferences, accounts, order history and alerts that make returning worth it. The Verge, for example, lets readers follow authors and topics to shape their homepage and a daily email digest.
- Community and access Forums, expert Q&A, workshops and events that offer real human connection.
- Access for agents The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to outside tools and data. A business that exposes its catalog, availability or data this way can become the source an agent uses to finish a task, such as checking stock or comparing options.
Gartner’s advice to brands points the same way: pursue discoverability across many channels, optimizing for both AI overviews and traditional SEO. These are the places where being chosen turns into a relationship: an inquiry, an account, a purchase or a signup.
Measuring it
If AI search is a brand channel, measure it like one. Traffic from AI answers is worth watching. Make it the goal, though, and teams start optimizing for the wrong things. A better scorecard follows the three stages.
| Stage | Question it answers | Example measures |
|---|---|---|
| Be seen | Are we showing up in AI answers? | Share of AI answers that mention or cite us for our core topics; how often assistants recommend us for high-intent questions; branded search volume over time |
| Be believed | Do people trust what they find? | How assistants describe our brand, products and people; accuracy of that description; engagement with reviews, methods and evidence on the page |
| Be chosen | Do people come to us directly? | Direct and branded visits; tool and feature usage; inquiries, signups, accounts and purchases |
Branded search is the closest thing to a long-term health check. It shows whether visibility you can’t see directly is turning into people who look for you by name.
I’ll be honest: most of these measures are new, and the tools for tracking them are still immature. Running a fixed set of questions through the major assistants on a regular schedule is a practical place to start. It’s imperfect, but it gives you a baseline and a trend.
Where to start
This is a lot of work, and it sequences well. Here’s how I’d spread it over a year:
- First quarter: foundations Audit your entity data across every property you own and fix inconsistencies. Add Organization or LocalBusiness schema with
sameAslinks, review your AI crawler policy, and build proper about and author pages. Start a baseline for how often you appear in AI answers. - Second quarter: content Restructure key pages around the questions people ask, with direct answers up top. Add sources, data, reviews and methods inline. Test formats beyond text, such as short video and visual explainers.
- Second half: utility and measurement Launch tools and features built on what only you have, and explore giving AI agents structured access to your data. Move reporting toward the seen, believed, chosen scorecard, with traffic as one input among several.
The work also crosses team lines. Entity consistency depends on marketing and communications. Schema and crawler policy sit with SEO and engineering. Tools and data access need product. On a small team, that may all be one or two people, which makes a shared scorecard even more useful.
Clicks are a shrinking share of how people discover things. The influence is still there to win. The sites that do well will show up in the answer, back it up with evidence, and give people a reason to come find them.
Sources
- Bain & Company, “Goodbye clicks, hello AI: Zero-click search redefines marketing” (Feb. 2025)
- Gartner, “Gartner Survey Finds 53% of Consumers Distrust AI-Powered Search Results” (Sept. 2025)
- Exploding Topics, “The AI Trust Gap: 82% Are Skeptical, Yet Only 8% Always Check Sources” (June 2025)
- Gallup/Knight Foundation, “American Views 2022: Part 2, Trust, Media and Democracy” (Feb. 2023)
- Roger C. Mayer, James H. Davis and F. David Schoorman, “An Integrative Model of Organizational Trust,” Academy of Management Review (1995)
- Dana Mahr et al., “Science journalists and public trust: comparative insights from Germany, Italy, and Lithuania”, Journal of Science Communication (Sept. 2025)
- Google Search Central, “Creating helpful, reliable, people-first content”
- Harry Clarkson-Bennett, “Entities Are The Foundation Of Modern SEO”, Search Engine Journal (Sept. 2026)
- Search Engine Land, “Google’s great clarity cleanup” (2025)
- Burt Herman, “Forget ‘Google Zero.’ We need to talk about ‘People Zero.’”, Nieman Lab (Dec. 2025)
- Dan Oshinsky, “How the 1/5/10 Strategy Can Improve Your Daily Newsletter”, Inbox Collective
- Nieman Lab, “The Verge is introducing personalized homepage feeds and email digests” (July 2025)
- Next Gen News, “Substantiate Mode”
- Model Context Protocol, modelcontextprotocol.io