AI hunt is changing astatine a gait nary of america has knowledgeable earlier successful marketing.
The presentations I saw astatine Zero Click NY highlighted some really overmuch AI hunt has changed complete the past six months and the characteristics that whitethorn go lasting features of the landscape.
Of each the points covered, these 7 stood retired arsenic the astir important.
From the emergence of the trading engineer, to the differences betwixt Claude and ChatGPT results, to Claude’s meteoric emergence among businesses complete the past 12 months, present are the astir impactful takeaways I near with.
1. Every AI relies connected different content
ChatGPT and Claude stock only about 8% of their citations, per Profound data. Put differently, 92% of what ChatGPT cites wouldn’t beryllium cited by Claude for the aforesaid query. A marque tin ain visibility successful 1 motor and beryllium virtually invisible successful the other.
On apical of that, they don’t conscionable mention different websites. They for illustration different kinds of content.
- ChatGPT indexes heavy connected organization content: Reddit, Quora, and forums dress up astir 16% of its citations.
- Claude sits astatine little than 1%. Claude, by contrast, loves listicles (36% of citations vs. ChatGPT’s ~20%) and sentiment contented (13.2% vs. 7.2%).
The narration to accepted hunt splits the aforesaid way. About 64% of the websites Claude cites besides look successful Google’s apical 50 for the aforesaid query. For ChatGPT, it’s only 37%.
In different words, “just do the SEO activity you’ve been doing” mightiness activity for Claude visibility, but apt won’t for ChatGPT.
Takeaway: It’s captious to pass to stakeholders that “AI visibility” will inevitably alteration by LLM, and you’ll person to prioritize them depending connected whom you’re trying to scope (more connected that later).
Track visibility by motor because the activity that wins successful 1 mightiness do almost thing successful another. UGC and organization seeding move ChatGPT, while listicles and accepted rankings move the needle connected Claude.
See wherever your marque appears successful AI search, wherever competitors are winning, and what it takes to go the reply AI recommends.
2. Claude is softly winning B2B — truthful series your optimization by audience
If you’ve seen the generative AI traffic-share charts, Claude looks for illustration a rounding error.
But web postulation is the incorrect chart. Roughly 85% of Anthropic’s gross comes from endeavor and API usage that ne'er shows up successful user postulation data.
The correct floor plan comes from Ramp’s AI Index, which tracks firm paper walk crossed tens of thousands of businesses.
A twelvemonth ago, azygous digits of those businesses were paying for Anthropic. Today, it’s 34.4% — up of OpenAI astatine 32.3%. For the first time, more businesses salary Anthropic than OpenAI.
I came distant from this position asking myself: If business users are progressively surviving successful Claude while consumers unrecorded successful ChatGPT, shouldn’t your optimization priorities attraction connected wherever your assemblage is?
Should B2B brands prioritize Claude visibility first? Should B2C brands prioritize ChatGPT first?
Almost cipher is doing this because group aren’t really reasoning astir who uses ChatGPT, Gemini, aliases Claude. That will apt change.
3. ChatGPT ads are here, and this is what we’re seeing
The infinitesimal is here: Your competitors are buying visibility done ChatGPT ads. ChatGPT ads are unrecorded and self-serve, sitting straight wrong the chat product.
The aforesaid 2 weeks brought GPT 5.5, citation chips turning into clickable hyperlinks (referral postulation jumped roughly 60% overnight, pinch homepage referral stock leaping from astir 3.5% to 24%), and Google moving AI Mode into its main hunt box.
None of that was an accident. The hyperlinks are the click-tracking rails an ads business needs. The study of much than 100,000 advertisement placements surfaced 3 things everyone should internalize.
ChatGPT Ads lucifer connected topic
Ads lucifer connected taxable similarity, not intent. Only 14% of existent personification prompts transportation commercialized intent, but 20% of prompts trigger ads — a mathematics problem tin service an ad.
The embedding study recovered that advertisement titles and descriptions are the azygous biggest drivers of which conversations you show up in. Your title and explanation are now targeting parameters, not conscionable creative.
Paying for ads
“Pay-to-play” is here. About 1 successful 5 advertisement placements appears against a mention of a nonstop competitor, and the marque mentioned organically shows up arsenic the advertiser only astir 8% of the time.
Someone other is doubly arsenic apt to beryllium the advertiser connected your integrated mention arsenic you are.
Startup CRM Adia is already placing ads against prompts wherever Salesforce appears, and Salesforce is playing defense, showing paid placements 40% of the time, moreover erstwhile it’s already mentioned organically.
Ad inventory is scarce and expensive
ChatGPT shows astir 1 advertisement per conversation, the median speech is 3 turns, only 30% of eligible users spot ads astatine all, and CPMs/CPCs are moving astir 4 times Meta’s.
Expect that to alteration successful predictable ways: much advertisement slots per answer, ads deeper into conversations, and follow-up suggestions engineered to create much turns, which intends much inventory.
The lesson: Organic AEO and paid defense are now the aforesaid job. If you’re search your brand’s integrated citations but not who’s advertizing against them, you’re seeing half the board.
Get the newsletter hunt marketers trust on.
See terms.
4. Claude is the astir straight optimizable AI correct now
When Claude searches the web, it pulls from Brave. Not “influenced by” Brave. According to the talk I saw, it pulls straight from it.
In Profound’s latest testing, 79.2% of Claude’s citations came straight from Brave’s apical 10 results for the balanced search.
There’s nary meaningful reshuffling aliases reranking. No different exemplary trusts its hunt supplier to thing for illustration this.
That makes Claude the astir straight optimizable exemplary successful AI search: a visible index, a checkable ranking, and (as we’ll spot next) predictable retrieval behavior.
If takeaway 2 convinced you that Claude matters for B2B, this is the playbook: Figure retired wherever you rank connected Brave for your cardinal prompts and dainty that arsenic your Claude visibility roadmap.
A model this transparent doesn’t enactment open. Optimize for it while it exists.
Dig deeper. Claude visibility whitethorn dangle heavy connected Brave Search rankings, caller information suggests
5. Claude only performs web searches a 3rd of the time
There’s a catch, and it’s a large one. ChatGPT triggers web hunt connected astir 95% of prompts. Claude searches only astir a 3rd of the clip — apt because each hunt costs money (Brave’s nationalist API pricing runs astir $5 per 1000 searches), truthful Claude has a existent financial inducement to reply from its weights.
You tin only optimize Claude erstwhile it really retrieves.
The bully news is that its hunt behaviour is predictable. Recency-framed prompts (“best X successful 2026”) trigger hunt astir 81% of the time.
Ranking-oriented prompts (“top 10…”) trigger it 67% of the time, location-dependent prompts 55%, and comparisons 51%.
Definitional and procedural prompts — “what is simply a CRM?” and “how do I…” — mostly don’t trigger hunt astatine all, which makes them astir worthless optimization targets for Claude.
The lesson: Before you put successful Claude visibility for a punctual category, trial whether Claude really searches for it.
Recency, rankings, locations, and comparisons are the aboveground areas wherever Brave rankings construe into Claude citations.
Everything other is answered from representation you can’t touch.
6. Query fan-out: A raffle connected 1 stage, near-deterministic connected another
Two speakers described the aforesaid system successful almost other terms, and the hostility betwixt them is instructive.
Query fan-out is the group of synthetic queries an AI motor runs successful the inheritance to stitchery contented earlier generating an answer.
Mike King of iPullRank framed it arsenic a raffle: You can’t spot aliases power the fan-out, truthful the occupation is to maximize your raffle tickets — much aboveground area crossed owned, earned, and shared properties, and, crucially, the correct contented formats.
Even if you rank for a fanned-out query, the incorrect format makes you ineligible.
His investigation points to caller measures of what wins retrieval — content-to-query cosine similarity and accusation summation some correlate powerfully pinch AI hunt performance.
Josh Blyskal of Profound’s information tells a different communicative for Claude specifically: Its fan-outs are near-deterministic.
The aforesaid punctual produces the aforesaid fan-out drawstring astir 65% of the time, and 94% of Claude’s fan-outs are stamped pinch the existent twelvemonth (ChatGPT does this only 17% of the time).
ChatGPT’s fan-outs churn constantly. Claude hardly moves. Both views whitethorn beryllium correct — for different engines.
Where fan-outs are stable, arsenic successful Claude, you tin publication them and build contented targeted straight astatine them. The year-stamping behaviour unsocial argues for putting the existent twelvemonth successful your titles.
Where fan-outs are volatile, arsenic successful ChatGPT, King’s raffle logic applies: Buy much tickets done formats and aboveground area.
One mechanism, 2 strategies, chosen per engine. Which, again, whitethorn require you to prioritize 1 complete the other.
7. The trading technologist is here, and agents are the caller workforce
It would beryllium easy to disregard “marketing engineer” arsenic a vendor-manufactured occupation title. The hiring marketplace says otherwise.
Google has hired its first trading engineer. Figma posted the domiciled astatine a $295,000 guidelines salary. RBC and Autodesk person made hires.
It became a breakout hunt word connected Google, and Google’s ain AI trading lead called trading engineers “the prosecute for 2026.”
Who is the perfect campaigner to go a trading engineer? Is this a domiciled wherever you commencement pinch an technologist and thatch them marketing, aliases vice versa?
The emerging statement floor plan is simply a marketer first — personification pinch transmission acquisition and sensation — who builds and maintains AI systems, reports to the caput of marketing, and unblocks the remainder of the team. A marketer who ships systems extremity to end.
The underlying logic is that astir trading activity decomposes into pipelines: extract data, toggle shape it, and load it location useful. Agents tin now tally those pipelines connected a loop.
- Monitoring competitor pricing and auto-generating income conflict cards.
- Watching landing pages and AEO beingness connected a schedule and staging A/B tests.
- Pulling objection themes retired of 800 income calls and drafting contented to reside each one.
Tasks that utilized to beryllium “we’ll get to it someday” projects go an day of supplier building. The constraint stops being headcount and becomes creativity.
If your squad doesn’t person personification successful this domiciled yet, there’s a bully chance it will eventually.
Track your visibility crossed AI search, uncover missed opportunities, and turn your beingness wherever customers are asking questions.
The occupation now: Figuring retired really this each works
There still is nary clear playbook for AI search. When that playbook does emerge, however, the first measurement whitethorn beryllium to prioritize 1 LLM complete different based connected who you want to find you.
And successful galore cases, that “who” is going to beryllium an agent. At the aforesaid time, we’ll person agents assisting america successful the activity we’re doing, and the request for group who tin technologist these systems will proceed to grow.
.png "How Kv Caching Slashes Llm Inference Costs At Scale")
English (US) · 
Indonesian (ID) ·