The real story behind all these AI headlines
Strip away the hype and the press releases and there’s one pattern running through almost every headline you just saw:
The surfaces where people discover brands are being rebuilt around AI systems, not around publishers, search results, or feeds.
Google is rolling out AI Overviews and “AI Mode.” Meta is wiring Andromeda and GEM into its ad stack and betting its capex on AI shopping. Search per user is already down nearly 20% in the U.S. while AI answers, AI images, AI agents, and AI-powered feeds absorb more intent and attention.
For CMOs, performance marketers, and media buyers, this is not a “keep an eye on it” moment. It is a “rewrite how we think about demand, attribution, and content” moment.
The shift: from query-based to system-based discovery
Historically, you could map demand to explicit user actions:
- Search query → ad impression → click → site visit → conversion
- Follow / subscribe → feed impression → click → conversion
Now, discovery is increasingly system-led:
- AI Overview answers the question without a click.
- Meta’s AI shopping surfaces products inside chat, feed, and agents.
- AI voice agents and assistants “choose” which brand to recommend.
- Semantic search and recommendation models decide what content “represents” a topic.
The user still has intent. You just see less of it. And the system is now the editor.
That’s the issue that matters: your growth now depends less on ranking for keywords or hacking individual algorithms, and more on being the preferred answer in a handful of dominant AI systems.
The three AI surfaces that actually matter
You do not need a strategy for “AI” in general. You need a strategy for three specific AI surfaces:
1. AI search surfaces (Google, Bing, vertical search)
Headlines about AI Overviews, semantic search, and “how to rank in AI Overviews” are all pointing at the same thing: classic SEO is being absorbed into AI-driven information retrieval.
The practical implications:
- Less click data, more inferred impact. Google is already hiding parts of AI Overview traffic. Search per user is declining. You will have to infer impact from blended metrics (brand search, direct, assisted conversions) instead of clean last-click SEO charts.
- Semantic coverage beats keyword checklists. Models care about topic completeness, entity relationships, and disambiguation, not whether you hit an exact phrase 7 times. You’re training a model, not a keyword density tool.
- Citations and mentions become the new “position 1.” Being named, cited, or visually featured in an AI answer is the new ranking. That means your brand and content need to be the canonical reference for specific problems, not just one more blue link.
2. AI ad systems (Meta Andromeda/GEM, Google PMax, TikTok automation)
Meta’s Andromeda and GEM, Google’s Performance Max, TikTok’s smart bidding – all the same story: media buying is being abstracted away into black boxes that eat signals and spit out results.
The practical implications:
- Audience targeting is commoditized. Your “secret” audience targeting is now table stakes. The system can find lookalikes and intent better than your manual segments in most cases.
- Creative and conversion data are your only real levers. What you feed these systems – assets, copy, product data, post-click experience, and conversion events – matters more than which lookalike you picked.
- Measurement is probabilistic, not forensic. You will not get clean, channel-by-channel truth. You’ll get modelled conversions, incrementality tests, and directional lift studies. That’s the new normal.
3. AI-native recommendation surfaces (feeds, agents, and assistants)
Threads, Pinterest, TikTok, Instagram comments, AI shopping agents, AI CRM – all of these are moving toward “the system decides what you see and who you hear from.”
The practical implications:
- Engagement quality becomes a ranking signal. Replying to Instagram comments boosts engagement by 21% because it teaches the system “this account is worth showing.” Same logic will apply to AI agents that track satisfaction and repeat usage.
- Trust and safety scores quietly gate your reach. “Does AI trust you?” is not a philosophical question. It’s a question about your content quality, complaint rates, spam signals, and brand reputation.
- Influencer and UGC signals bleed into search and AI answers. Influencer marketing and SEO are converging. If creators and users talk about you, models treat you as a more likely answer.
What this breaks in your current strategy
If you keep operating like it’s 2019, three things will quietly erode:
- Your comfort with channel-level ROI. Clean “search did X, social did Y” narratives are going to get noisier as AI surfaces intercept more demand before it hits your site.
- Your content operating model. Churning out long-tail SEO posts and generic social content will not train AI systems to see you as an authority. It will just add to the noise they compress.
- Your negotiating power with platforms. As AI systems intermediate more of the experience, you become more dependent on their opaque rules. If you’re not modelling incrementality and building owned demand, you’re flying blind.
A practical operating system for the AI surface era
You don’t need a moonshot. You need a clear operating system that recognizes the new physics. Here’s a pragmatic way to run it.
1. Redefine “search” as “answer share”
Stop obsessing over blue-link rankings and start measuring your share of answers.
- Map your critical questions. For each product line, list the 20-50 questions that actually drive revenue: “best X for Y,” “how to solve Z,” “alternatives to A,” “pricing for B.”
- Audit AI surfaces, not just SERPs. For those questions, check:
- AI Overviews (where available)
- Featured snippets and People Also Ask
- YouTube search and suggested videos
- Reddit, Quora, and major forums where AI models train
- Score your presence. For each question, score:
- Are we cited, mentioned, or visually present in AI answers?
- Do we own any canonical guides / tools / data on this?
- Do creators mention us when answering this?
Your goal: increase your “answer share” on the questions that actually move revenue, not on vanity traffic.
2. Build content that trains models, not just ranks pages
AI systems are pattern recognizers. Treat your content as training data.
- Go deep on entities and relationships. Create content that fully explains a topic, the entities involved (brands, tools, use cases), and how they relate. Think structured, clear, and unambiguous – like you’re writing for an engineer, not a skimmer.
- Use consistent, descriptive language. If you call your product five different things across your site, don’t be surprised when models get confused. Pick canonical terms and stick to them.
- Publish reference assets, not just blog posts. Calculators, benchmarks, glossaries, data studies, and how-to frameworks tend to be reused, cited, and scraped. That makes you a more likely citation source.
3. Treat AI ad systems as partners you have to feed
You can’t see inside Meta’s Andromeda or Google’s PMax, but you can control what you give them.
- Upgrade your conversion plumbing. Invest in clean, privacy-compliant conversion tracking and server-side events. Bad data in means bad optimization out.
- Design creative for exploration, not just exploitation. Give systems a diverse set of creative angles, hooks, and formats so they can find pockets of performance. One “hero” ad and a few resized versions is a 2018 move.
- Run structured experiments. Use geo splits, audience holdouts, and time-based tests to measure incrementality at a campaign or channel level. Stop expecting the platform to hand you truth.
4. Build an “AI trust profile” for your brand
AI systems are constantly asking: “Is this source safe, useful, and satisfying?” You should be asking the same thing about how you appear to them.
- Audit risk signals. High complaint rates, spammy email practices, misleading claims, and thin content all drag down how systems treat you. Clean this up now; it compounds.
- Standardize your factual spine. Make sure your pricing, features, specs, and key facts are consistent across your site, docs, app, feeds, and major directories. Inconsistent truth is poison for models.
- Instrument satisfaction. Track CSAT, NPS, refund rates, and support resolution times and connect them (where possible) to your ad and CRM systems. Over time, these will correlate with how AI agents choose who to recommend.
5. Shift your measurement culture from precision to decision
The AI surface era will not give you perfect attribution. You need “good enough to decide.”
- Define a small set of “control metrics.” For example: blended CAC, payback period, new customer growth, repeat purchase rate, and brand search volume. Make these the scoreboard.
- Accept modelled truth where it’s the best you can get. MMM, incrementality tests, and platform lift studies are imperfect. They’re also the only way to see the forest when user-level data is obscured.
- Train teams to live with ambiguity. Replace “prove this channel works” with “what evidence would change our spend decision?” and design tests around that.
Where to reallocate attention and budget in the next 12-18 months
If you’re a CMO or performance leader, here’s where to bias your next planning cycle:
- Upweight:
- Deep, reference-grade content tied to revenue-driving questions.
- Creative production and testing capacity for AI-optimized ad systems.
- Measurement infrastructure: clean data, incrementality testing, and basic MMM.
- Brand and trust work that shows up in how models perceive you (PR, reviews, creator partnerships, support quality).
- Downweight:
- Manual micro-optimizations of targeting that AI already does better.
- High-volume, low-signal content created “for SEO” or “for the feed.”
- Attribution battles that chase false precision instead of better decisions.
The platforms have already picked their path: AI systems will sit between your brand and your customers across search, social, and shopping. You can either keep optimizing for the old surfaces, or you can start operating for the new reality: you are now marketing to people and to the models that decide what those people see.