Rebuilding a session you have delivered twice before is humbling, and I had to rebuild this one almost completely, because the thing it describes refused to hold still. Search is shifting from links to answers, and a buyer who gets one answer never sees the nine links underneath it. Either your content is inside that answer or the buyer does not know you exist.

Mine was the last session of the conference, the slot where people are usually thinking about flights and checkout times, and the room was full. Thank you to everyone who stayed. You made it a conversation rather than a presentation, and the questions at the end were some of the best I have gotten on this topic.

The room was a mix of technology leaders and marketers, which was the point. The problem I wanted them to leave with belongs to both of them, and most companies have assigned it to neither.

Last year the anxiety in every conversation was zero-click search. Organic traffic was falling and people were rehearsing how to explain a thirty percent decline to their board. This year I asked them to look at a different number. Teams that report on conversion instead of traffic are seeing the opposite story. Adobe shared data at Summit showing conversion swinging from deeply negative to strongly positive among AI-referred visitors, and the explanation is simple once you see it. Buyers are doing their research inside ChatGPT, Claude, and Google's AI Overviews before they ever reach your site, so the visitor who finally lands on your page has already finished the comparison work your funnel was built to walk them through. The practical move is to stop re-educating them. Put the fastest possible path to a salesperson or a purchase right up front, because they arrived closer to a decision than your analytics suggest.

The old benchmarks confirm the shift from the other direction. I used to advise clients to aim for a page and a half and ninety seconds per session. Today the majority of traffic on most sites spends under nineteen seconds, and that is not spam. It is machines reading on behalf of humans who will never visit.

So the question becomes whether the machines can read you at all.

This is where the session turned, because the honest answer for many companies in that room was no, and not because their content is weak. The blockers are almost always in the plumbing. I started noticing it in my own workflow. I would drop a press release URL into an LLM and get told the page could not be accessed. When I dug into why, the same culprits kept appearing: a robots.txt file written twenty years ago for a world that no longer exists, or a website built entirely in JavaScript that looks polished to a person and arrives as a blank page to a crawler. The third culprit was the one that surprised people most, settings at the CDN layer that nobody in marketing knew existed.

Cloudflare is the example I used, and I want to be careful with it because the policy itself is sensible. Cloudflare now lets publishers decide which AI crawlers get access and on what terms, which is exactly right for a media company protecting paid content. What turns a sensible policy into a findability problem is that the people who configure it and the people who feel its effects sit in different departments. IT and security own the setting, marketing inherits the result, and in most companies those two teams have never discussed it. A gentleman sitting next to me on the flight in checked his own properties when we landed and called me to say I was right, a long list of AI crawlers was being managed at a layer his marketing team had never looked at.

That gap has a strange downstream effect. When an AI cannot reach your current content, it cites whatever it can reach, which is how a fifteen-year-old article about your company ends up as the top citation while your own site is invisible. Models are hallucinating less these days partly because they lean harder on third-party content they treat as neutral, comparison posts from bloggers, coverage from the press. If those sources answer the buyer's question vividly, they get the citation that should have been yours.

Open robots.txt to the crawlers you want, including the ones your network controls quietly exclude. Get your key content into raw HTML. Then make it intelligible with schema.org markup, an Article type for content pages, Person for your authors and executives, FAQ for product and support pages. The schema gap I see most often in audits of global brands is region, where missing area-served markup convinces an agent the company only operates in Europe, and an entire continent of buyers never sees them in an answer.

The last fix is the one I pushed hardest, because it is the cheapest and the most neglected. Put real people on your content. People do not search for the world's best software anymore, they ask who the visionary behind it is, and if that person works for you but has no author page an AI can verify, you have hidden your best credibility signal from the systems deciding whether to cite you. The internal objection is always speed, content approval takes two weeks and the experts are busy. Fine. Have them talk for fifteen minutes into a recorder and let a writer shape it as a co-author. The era of hiring someone to push out forty posts a week is over anyway. HubSpot taught a generation that volume wins, and that lesson has expired.

View source and see if your text is actually there, then open your site in reader mode, which is a decent proxy for what an AI parser sees. A page that comes up blank in reader mode is giving the machines nothing to work with.

One of the questions asked at the end was what to do when an AI gets your company wrong. Interrogate it. Ask why it recommended what it did. The answer will usually point you straight at the content, or the missing content, that taught it the error.

The homework I gave the room for the flight home was to ask an AI who the leading experts in their category are, and sit with whatever comes back. The harder homework starts after landing: find whoever owns your CDN settings and have the conversation that IT and marketing should have had a year ago. A blueprint on staying relevant in AI engines is available from Info-Tech Research Group for anyone who wants the full framework.

Every fix in this session is cheap. What is expensive is the absence of an owner.