You Can't Optimize for a Browser. You Optimize for the Assistant.

elisabeth hitz · july 19, 2026 · 6 min read

There is no such thing as AI browser optimization. A browser has no ranking function, no index, and no opinion about you. The thing forming an opinion about you is the model, and the discipline of getting named by a model is different enough from SEO that treating it as "SEO but for AI" is why most brands are invisible.

OpenAI proved the point by announcing on July 9, 2026 that it is retiring its own browser. Atlas goes dark on August 9, and the capability folds into ChatGPT. Ten months, done. If your visibility plan had a line item for AI browsers, that line item was never real. The full breakdown of what happened is here.

Here is what to optimize instead.

What is the difference between SEO and AEO?

SEO optimizes a page so a human clicks it. AEO, answer engine optimization, optimizes a page so a model quotes it. The overlap is real but partial, and the non-overlapping part is where every opportunity currently sits.

That last cell is the one to sit with. You can win from an AI answer without ever receiving a visitor. Brands named in AI Overviews earn meaningfully more organic clicks than brands that are not, even accounting for the click suppression. The mention itself does work.

Why is my content not being cited by ChatGPT?

Usually one of five reasons. In order of how often I see it:

1. Your answer is not extractable. Models lift self-contained paragraphs. If your post opens with three paragraphs of context before the answer, there is nothing to lift. Answer the headline question in the first 40 words, in a paragraph that makes sense with nothing around it. This single change moves more citations than any other fix.

2. Your content is generic enough to paraphrase away. A model reading five sources that all say "consistency matters in branding" writes one sentence and cites nobody. A model reading a source that says organic CTR on informational queries dropped from roughly 1.76% to 0.61% has to say whose number that is, because the claim cannot be restated without its origin. Same with a named outcome: one creator I work with, Ty Murph, moved from $150 to $800 on the same category of brand deal. You cannot paraphrase that into a generic sentence, the name and the numbers are the claim. Specificity forces attribution. Generic expertise is invisible expertise.

3. All your evidence is on your own domain. AI engines cite earned, third-party sources far more than brand-owned pages. Your About page is a claim. A podcast transcript, a customer's write-up, a forum answer with your name on it, a review-site profile, those are evidence. If every mention of you on the internet is published by you, models have no corroboration and will route around you.

4. Your structure is unreadable to a machine. No schema, headings that say "Our Approach" instead of the question a person would type, no tables, no lists, no dates. You are making the model work, and there are ten competitors making it easy.

5. You are stale. Recency is weighted far more heavily in AI answers than in blue links. A page updated this month outranks a definitive page from 2024. Most sites publish and abandon. Maintaining ten URLs beats publishing forty.

How do I actually get cited? The mechanics.

Six things, roughly in ROI order. None require rebuilding your site.

1. 1. Front-load the answer. First 40 words, self-contained, declarative. No throat-clearing.
2. 2. Question-shaped H2s. Write the heading as the exact sentence a human would type or say. "Why is my content not being cited by ChatGPT?" not "Citation Dynamics."
3. 3. Ship extractable blocks. Every page: at least one comparison table, one numbered list, one bolded standalone definition. These are the units models copy verbatim.
4. 4. Add Article + FAQPage schema, with datePublished and dateModified. Fifteen minutes. Most competitors skip it.
5. 5. Publish llms.txt at your root, listing your key pages with one-line descriptions. Cheap, uncontested, still rare.
6. 6. Go earn mentions off-domain. Three guest posts, two podcasts, and answering ten real questions in a forum where your buyers live will do more for AI citation than twenty blog posts on your own site.

How do I measure whether it worked?

Two numbers. Not vanity metrics.

Citation rate. Once a week, ask the same five buying-intent questions across ChatGPT, Claude, Perplexity, and Google AI Mode. Log whether you are named. That is the scoreboard. It is manual, it takes ten minutes, and almost nobody does it, which is exactly why it is an edge.

AI-referred conversions. Build a GA4 custom channel group capturing chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com. Without it, these land in "direct" and you will conclude nothing is working.

Expect low volume and high quality. AI referral traffic is 95-96% smaller than Google search traffic per TollBit's measurement, and converts materially better, because the assistant already did the filtering, comparison, and objection handling before the person clicked.

You are not replacing your traffic. You are replacing your traffic with a smaller, warmer stream and accepting that the volume is gone. Businesses that fight that fact spend two years losing. Businesses that accept it rebuild around it in one quarter.

What this means strategically

If assistants read your content and hand the answer to your buyer, then content that can be fully summarized is content that has been fully substituted. The how-to guide, the listicle, the definitional post, those are exactly what models compress best. HubSpot reportedly lost 70-80% of its organic traffic on precisely that content shape.

What survives summarization: proprietary data, a delivered artifact, a system somebody installs and runs, a diagnosis of their specific situation. Files, not chats. Infrastructure, not tools.

That is the same conclusion OpenAI reached about its own browser: the wrapper is disposable, the capability is the asset. I unpacked the whole pattern in the assistant is the new homepage.

or just follow along. new field notes most weeks on x, instagram, and tiktok.