What the heck is going on with AI search?

Last week I covered part one of what’s happening in AI search.

How ChatGPT retrieves more than it cites.
Why SEO still matters.
Why headings and query match are doing more work than backlinks.
Why bigger pages are not always better.
Why your website is only part of the visibility picture.

This week is part two. Where I’ll be covering:

When does ChatGPT actually search the web?
Can AI systems access your site in the first place?
Are you measuring the impact properly?
Does visibility actually create traffic?
If your content gets cited, does your brand get the value?

You can check out last weeks newsletter here.

Semrush analyzed 17 months of clickstream data and found that, as of February 2026, ChatGPT enabled web search on only 34.5% of queries. That’s down from 46% in late 2024.

It also found ChatGPT referral traffic to the wider web grew 206% in 2025, but more than 30% of referrals went to just 10 domains, with Google alone taking 21.6%.

The problem is that a lot of AI search advice assumes every answer is built from live web results. It’s not.

Sometimes ChatGPT searches. Sometimes it pulls answers from what it already knows. When it does send traffic, that traffic is concentrated.

So the action here is to split AI visibility into two separate focus points.

1: Search-triggered visibility. Where ChatGPT is more likely to browse: recent information, source requests, comparison checks, pricing changes, product updates, news, and anything the model is uncertain about.

For those topics, you need pages that are crawlable, current, clearly dated, and easy to cite.

2: Non-search visibility. Where ChatGPT pulls from existing model knowledge instead of live sources. That’s harder to influence with one new page.

For those topics, you need repeated brand-category signals over time: the same positioning, use cases, product descriptions, and proof points showing up across your site and the web.

This is one of the challenges with AI search.

If ChatGPT searches, your page needs to be findable.
If it doesn’t search, your brand needs to already be understood by the model.

Peec’s 2026 AI search guide is an example of why traffic alone isn’t enough.

For example, Chime appeared in 66% of AI chats in the U.S., while Revolut appeared in 33%. This is what marketers should be tracking: how often a brand appears in relevant AI answers.

The issue is that LLM traffic is underreported because many AI-influenced journeys have no click to track.

A buyer might see your brand in ChatGPT, search for you on Google, then come through organic or direct. GA credits the last visible channel, not the AI answer that informed the decision. 

So the mistake is reporting AI search like normal SEO traffic.

Instead, it’s better to track four things:

Visibility: how often you appear.
Position: how high you appear in the answer.
Sentiment: what AI says about you.
Revenue influence: whether buyers say AI helped them find or evaluate you.

But this data doesn’t come from Google Analytics. It comes from testing the AI answers by using prompts.

That doesn’t mean testing one random prompt, as it’s too unreliable. One answer might include you. The next might not. One model might recommend you. Another might ignore you.

So it has to be built around groups of prompts that represent influence for the user and brand:

Category prompts: ‘best tools for X’
Comparison prompts: ‘X vs Y’
Use-case prompts: ‘best platform for a small B2B team doing X’
Problem prompts: ‘how to solve X without hiring Y’
Bottom-funnel prompts: ‘is X worth it for a company like mine?’

Then track visibility, position, and sentiment across the whole group. That gives you a more useful understanding of AI search performance.

Not ‘did we show up once?’

But: ‘Are we consistently visible when users ask this type of question?’

That’s what you can compare against branded search, direct traffic, demo notes, sales calls, and self-reported attribution.

Also, add AI tools as a source option on forms. Ask:

Did ChatGPT, Perplexity, Gemini, Claude, or Google AI Overviews help you find or evaluate us?

Because AI might be creating the demand, even when another channel gets the credit.

OtterlyAI analyzed 1+ million AI citations across ChatGPT, Perplexity, and Google AI Overviews from January to February 2026.

One of the biggest findings had nothing to do with writing.

It found 73% of sites had technical barriers blocking AI crawler access, including robots.txt blocks, CDN restrictions, and JavaScript rendering issues.

That’s a problem because some teams focus on rewriting and optimizing content when the real issue is access.

If AI systems cannot crawl or parse the page, the content doesn’t even get a look-in.

So before chasing another AI content tactic, check the basics:

Robots.txt.
CDN security rules.
JavaScript-rendered content.
Server logs.
AI crawler access for GPTBot, ChatGPT-User, OAI-Searchbot, ClaudeBot, and PerplexityBot.

If those are blocked, you are unintentionally creating a technical problem preventing you from showing in AI search.

SE Ranking’s latest AI Overview tracking shows how much the feature has evolved by category.

By February 2026, AI Overviews appeared most often in Finance at 78%, Business at 77%, Career and Jobs at 76%, Relationships at 74%, and Sports and Exercise at 72%. 

Finance had moved from 11% in May 2025 to 78% in February 2026, roughly a 7x increase.

That shows that AI Overviews are moving from simply a lifestyle or basic information feature into categories where people make decisions.

Money.
Work.
Careers.
Business.

Which means marketers need to adjust what they monitor.

It’s not enough to know where you rank. You need to know whether an AI Overview appears, what it says, which brands or sources it uses, and whether your page is being pushed below the visible part of the result.

A ranking can stay the same while the SERP around it changes completely. This is how visibility gets weaker without a drop in rankings.

Seer analyzed 541,213 LLM responses across 20 brands and found what it calls ‘ghost citations.’

That is when your content gets cited, but your brand doesn’t get mentioned, or a competitor gets the recommendation instead.

That’s a different kind of loss.

Your page helps with the answer, and someone else gets the brand value.

It’s not always an issue with content quality though. In some cases, the difference is brand entity strength. One brand is connected strongly enough to the category to be named. The other provides useful content but doesn’t get pulled into the answer.

As mentioned already, stop looking at citations as the final win.

Instead, think about whether you are cited, mentioned, and recommended.

If your URL is cited but your competitor is named, the page worked, but the brand didn’t.

That means you need stronger brand-category association: clearer product positioning, stronger comparison pages, more direct use-case language, better third-party reinforcement, and content that connects your brand name to the problems you want to own.

Ahrefs re-ran its AI Overview CTR study using December 2025 data and found that AI Overviews now correlate with a 58% lower average click-through rate for the top-ranking page.

The study used 300,000 keywords, split between 150,000 AI Overview keywords and 150,000 informational keywords without AI Overviews.

Which is frustrating because:

You can rank.
You can be visible.
You can still lose the click.

This doesn’t mean visibility is worthless. It means traffic is no longer the only outcome of search visibility.

AI Overviews can influence the buyer before they visit your site. They can answer enough of the query to stop the click. They can also make your brand familiar before someone searches again later.

So marketers need to look at more than rankings and traffic.

Look at branded search.
Look at direct traffic.
Look at assisted conversions.
Look at demo notes.
Look at sales calls.
Look at whether prospects mention ChatGPT, Gemini, Perplexity, or Google AI Overviews.

Because AI search might be influencing demand before it sends a visit.

Sometimes, it might not send the visit at all.

Part one showed how AI systems retrieve, filter, and cite information.

Part two shows what happens around that.

ChatGPT doesn’t always search the web.
AI journeys are difficult to attribute.
Technical access can block visibility before content is judged.
AI Overviews are moving into higher-value categories.
Citations don’t always turn into recommendations.
Visibility doesn’t guarantee traffic.

That’s what marketers need to focus on now, not just publishing more content. 

It’s cleaner technical access, stronger brand-category association, better attribution, and content that is built to be understood, cited, and connected back to the brand.