7 AI search shifts you can’t afford to ignore

AI search is altering at a tempo none of us has skilled earlier than in advertising.

The shows I noticed at Zero Click on NY highlighted each how a lot AI search has modified over the previous six months and the traits which will change into lasting options of the panorama.

Of all of the factors lined, these seven stood out as a very powerful.

From the rise of the advertising engineer, to the variations between Claude and ChatGPT outcomes, to Claude’s meteoric rise amongst companies over the previous 12 months, listed below are essentially the most impactful takeaways I left with.

1. Each AI depends on totally different content material

ChatGPT and Claude share solely about 8% of their citations, per Profound information. Put in a different way, 92% of what ChatGPT cites wouldn’t be cited by Claude for a similar question. A model can personal visibility in a single engine and be just about invisible within the different. 

On high of that, they don’t simply cite totally different web sites. They like totally different sorts of content material. 

• ChatGPT indexes closely on group content material: Reddit, Quora, and boards make up roughly 16% of its citations. 
• Claude sits at lower than 1%. Claude, in contrast, loves listicles (36% of citations vs. ChatGPT’s ~20%) and opinion content material (13.2% vs. 7.2%).

The connection to conventional search splits the identical approach. About 64% of the web sites Claude cites additionally seem in Google’s high 50 for a similar question. For ChatGPT, it’s solely 37%.

In different phrases, “simply do the web optimization work you’ve been doing” would possibly work for Claude visibility, however possible gained’t for ChatGPT.

Takeaway: It’s crucial to speak to stakeholders that “AI visibility” will inevitably range by LLM, and also you’ll should prioritize them relying on whom you’re making an attempt to achieve (extra on that later).

Observe visibility by engine as a result of the work that wins in a single would possibly do nearly nothing in one other. UGC and group seeding transfer ChatGPT, whereas listicles and conventional rankings transfer the needle on Claude.

2. Claude is quietly successful B2B — so sequence your optimization by viewers

If you happen to’ve seen the generative AI traffic-share charts, Claude appears to be like like a rounding error.

However internet visitors is the mistaken chart. Roughly 85% of Anthropic’s income comes from enterprise and API utilization that by no means exhibits up in shopper visitors information.

The appropriate chart comes from Ramp’s AI Index, which tracks company card spend throughout tens of 1000’s of companies.

A yr in the past, single digits of these companies had been paying for Anthropic. At the moment, it’s 34.4% — forward of OpenAI at 32.3%. For the primary time, more businesses pay Anthropic than OpenAI.

I got here away from this presentation asking myself: If enterprise customers are more and more residing in Claude whereas customers reside in ChatGPT, shouldn’t your optimization priorities give attention to the place your viewers is?

Ought to B2B manufacturers prioritize Claude visibility first? Ought to B2C manufacturers prioritize ChatGPT first?

Nearly no one is doing this as a result of folks aren’t actually fascinated about who makes use of ChatGPT, Gemini, or Claude. That may possible change.

3. ChatGPT advertisements are right here, and that is what we’re seeing

The second is right here: Your rivals are shopping for visibility via ChatGPT advertisements. ChatGPT advertisements are reside and self-serve, sitting instantly contained in the chat product.

The identical two weeks introduced GPT 5.5, quotation chips turning into clickable hyperlinks (referral visitors jumped roughly 60% in a single day, with homepage referral share leaping from roughly 3.5% to 24%), and Google shifting AI Mode into its essential search field.

None of that was an accident. The hyperlinks are the click-tracking rails an advertisements enterprise wants. The evaluation of greater than 100,000 advert placements surfaced three issues everybody ought to internalize.

ChatGPT Advertisements match on subject

Advertisements match on subject similarity, not intent. Solely 14% of actual person prompts carry industrial intent, however 20% of prompts set off advertisements — a math downside can serve an advert.

The embedding evaluation discovered that advert titles and descriptions are the one greatest drivers of which conversations you present up in. Your title and outline at the moment are concentrating on parameters, not simply artistic.

Paying for advertisements

“Pay-to-play” is right here. About one in 5 advert placements seems towards a point out of a direct competitor, and the model talked about organically exhibits up because the advertiser solely about 8% of the time.

Another person is twice as prone to be the advertiser in your natural point out as you’re.

Startup CRM Adia is already inserting advertisements towards prompts the place Salesforce seems, and Salesforce is enjoying protection, displaying paid placements 40% of the time, even when it’s already talked about organically.

Advert stock is scarce and costly

ChatGPT exhibits roughly one advert per dialog, the median dialog is three turns, solely 30% of eligible customers see advertisements in any respect, and CPMs/CPCs are working round 4 occasions Meta’s.

Count on that to vary in predictable methods: extra advert slots per reply, advertisements deeper into conversations, and follow-up ideas engineered to create extra turns, which suggests extra stock.

The lesson: Natural AEO and paid protection at the moment are the identical job. If you happen to’re monitoring your model’s natural citations however not who’s promoting towards them, you’re seeing half the board.

4. Claude is essentially the most instantly optimizable AI proper now

When Claude searches the online, it pulls from Brave. Not “influenced by” Courageous. In accordance with the discuss I noticed, it pulls instantly from it.

In Profound’s newest testing, 79.2% of Claude’s citations got here instantly from Courageous’s high 10 outcomes for the equal search.

There’s no significant reshuffling or reranking. No different mannequin trusts its search supplier to something like this.

That makes Claude essentially the most instantly optimizable mannequin in AI search: a visual index, a checkable rating, and (as we’ll see subsequent) predictable retrieval conduct.

If takeaway 2 satisfied you that Claude issues for B2B, that is the playbook: Determine the place you rank on Courageous on your key prompts and deal with that as your Claude visibility roadmap.

A window this clear doesn’t keep open. Optimize for it whereas it exists.

Dig deeper. Claude visibility may depend heavily on Brave Search rankings, new data suggests

5. Claude solely performs internet searches a 3rd of the time

There’s a catch, and it’s a giant one. ChatGPT triggers internet search on roughly 95% of prompts. Claude searches solely a couple of third of the time — possible as a result of each search prices cash (Courageous’s public API pricing runs round $5 per thousand searches), so Claude has an actual monetary incentive to reply from its weights.

You may solely optimize Claude when it truly retrieves.

The excellent news is that its search conduct is predictable. Recency-framed prompts (“greatest X in 2026”) set off search about 81% of the time.

Rating-oriented prompts (“high 10…”) set off it 67% of the time, location-dependent prompts 55%, and comparisons 51%.

Definitional and procedural prompts — “what’s a CRM?” and “how do I…” — largely don’t set off search in any respect, which makes them practically nugatory optimization targets for Claude.

The lesson: Earlier than you spend money on Claude visibility for a immediate class, check whether or not Claude truly searches for it.

Recency, rankings, places, and comparisons are the floor areas the place Courageous rankings translate into Claude citations.

Every thing else is answered from reminiscence you possibly can’t contact.

6. Question fan-out: A raffle on one stage, near-deterministic on one other

Two audio system described the identical mechanism in nearly reverse phrases, and the strain between them is instructive.

Query fan-out is the set of artificial queries an AI engine runs within the background to collect content material earlier than producing a solution.

Mike King of iPullRank framed it as a raffle: You may’t see or management the fan-out, so the job is to maximise your raffle tickets — extra floor space throughout owned, earned, and shared properties, and, crucially, the correct content material codecs.

Even if you happen to rank for a fanned-out question, the mistaken format makes you ineligible.

His analysis factors to new measures of what wins retrieval — content-to-query cosine similarity and knowledge achieve each correlate strongly with AI search efficiency.

Josh Blyskal of Profound’s information tells a distinct story for Claude particularly: Its fan-outs are near-deterministic.

The identical immediate produces the identical fan-out string about 65% of the time, and 94% of Claude’s fan-outs are stamped with the present yr (ChatGPT does this solely 17% of the time).

ChatGPT’s fan-outs churn continuously. Claude barely strikes. Each views could also be proper — for various engines.

The place fan-outs are secure, as in Claude, you possibly can learn them and construct content material focused instantly at them. The year-stamping conduct alone argues for placing the present yr in your titles.

The place fan-outs are unstable, as in ChatGPT, King’s raffle logic applies: Purchase extra tickets via codecs and floor space.

One mechanism, two methods, chosen per engine. Which, once more, might require you to prioritize one over the opposite.

7. The advertising engineer is right here, and brokers are the brand new workforce

It will be straightforward to dismiss “advertising engineer” as a vendor-manufactured job title. The hiring market says in any other case.

Google has employed its first advertising engineer. Figma posted the function at a $295,000 base wage. RBC and Autodesk have made hires.

It turned a breakout search time period on Google, and Google’s personal AI advertising lead known as advertising engineers “the rent for 2026.”

Who’s the perfect candidate to change into a advertising engineer? Is that this a job the place you begin with an engineer and educate them advertising, or vice versa?

The rising consensus profile is a marketer first — somebody with channel expertise and style — who builds and maintains AI methods, experiences to the pinnacle of selling, and unblocks the remainder of the group. A marketer who ships methods finish to finish.

The underlying logic is that the majority advertising work decomposes into pipelines: extract information, rework it, and cargo it someplace helpful. Brokers can now run these pipelines on a loop.

• Monitoring competitor pricing and auto-generating gross sales battle playing cards.
• Watching touchdown pages and AEO presence on a schedule and staging A/B exams.
• Pulling objection themes out of 800 gross sales calls and drafting content material to handle each.

Duties that was once “we’ll get to it sometime” initiatives change into a day of agent constructing. The constraint stops being headcount and turns into creativity.

In case your group doesn’t have somebody on this function but, there’s an excellent likelihood it is going to finally.

The job now: Determining how this all works

There nonetheless is not any clear playbook for AI search. When that playbook does emerge, nevertheless, step one could also be to prioritize one LLM over one other primarily based on who you need to discover you.

And in lots of instances, that “who” goes to be an agent. On the identical time, we’ll have brokers aiding us within the work we’re doing, and the demand for individuals who can engineer these methods will proceed to develop.