The real pattern in all these headlines
Ignore the AI hype words and look at what’s actually going on in those headlines:
- “Are AI Overviews Stealing Your Clicks?”
- “How AI Agents See Your Website”
- “What Pichai’s Interview Reveals About Google’s Search Direction”
- “AEO strategy for SaaS” (AEO = Answer Engine Optimization)
- “Future of Marketing Briefing: The ad industry has an AI label problem”
- “OpenAI has quietly launched its ads manager”
- “Why Pfizer and other blue-chip brands are building internal AI search hubs”
The throughline isn’t “AI” in the abstract. It’s intermediation.
Search engines, social feeds, and now AI agents are stepping between you and your customer, summarizing your content, compressing your brand, and deciding what gets surfaced. Your traffic, your ROAS, your “brand” – all are now downstream of systems that:
- Read your site and product feeds like a machine, not a human
- Answer the user directly instead of sending them to you
- Prefer structured, consistent, low-friction information
If you’re still optimizing for “10 blue links” and last-click ROAS, you’re playing a game that’s being decommissioned in real time.
The new problem: AI is the homepage, not your site
In the answer-engine era, three shifts matter more than any new ad format:
- AI overviews and agents are becoming the default interface. Users ask questions; systems answer. Clicks become a secondary outcome, not the primary one.
- Your content is raw material for someone else’s answer. Title tags, schema, product feeds, FAQs – these are no longer just SEO hygiene. They are the training data for the interface that sits between you and demand.
- Attribution is degrading. Misreported ROAS, “view-through” fantasy, AI assistants recommending products without traceable clicks – your measurement stack is being stress-tested.
The practical question for operators is not “What’s our AI strategy?” but:
How do we market in a world where intermediaries answer first and send traffic later – if at all?
Principle 1: Optimize for interpreters, not just visitors
You now have two audiences:
- Humans, with emotions, objections, and context
- Machines, with parsers, models, and ranking systems
Most brands still build for humans and hope machines “figure it out.” That’s how you end up in AI overviews as a generic bullet point instead of the recommended choice.
Make your site “agent-readable”
If an AI agent or overview panel had to answer “Who is this for?” and “What do they sell?” from your site in 2 seconds, could it?
Practical moves:
-
Brutally clear positioning, above the fold.
Replace “innovative platform for modern teams” with “Project management software for 20-500 person B2B teams.” -
Structured data everywhere it’s even remotely justified.
Product schema, FAQ schema, organization schema, how-to schema, event schema. This is not “SEO checklists”; it’s how you feed answer engines unambiguous facts. -
Canonical, non-cannibalized pages for key intents.
That Moz “Cannibalization” thread matters more now. If your own site sends mixed signals on which page is “the” answer, expect AI summaries to be just as muddled. -
Machine-friendly pricing and feature tables.
Clear tables, consistent labels, and explicit units (“$/seat/month”) make it easier for models to compare you fairly instead of hallucinating gaps.
Principle 2: Treat answer engines as a performance channel
Most teams talk about AI overviews like weather: “We’re seeing some impact.” That’s not a plan.
Map where AI is already intermeditating your funnel
Start with three simple audits:
-
Query audit.
For your top 50 non-brand queries (by revenue), check:- Does Google show AI overviews? How often?
- Are you referenced in those overviews? As what – example, source, recommendation?
- Are competitors being framed as “best for X” while you’re just a mention?
-
Page-type audit.
Which page types get surfaced or summarized?- Buying guides, comparisons, FAQs, docs, product pages?
- Are there gaps where third-party sites are “owning” the answer instead of you?
-
Attribution audit.
Where have you seen:- Stable search volume but falling clicks?
- Stable clicks but lower on-site engagement (users already “pre-answered”)?
- Brand search rising while generic search CTR falls?
Treat this like a new ad product: understand its inventory, behavior, and impact before you start reacting tactically.
Principle 3: Build for branded demand, not just generic discovery
As answer engines hoard generic intent (“best CRM for startups”), the most defensible asset becomes branded intent (“HubSpot pricing,” “Notion templates,” “Figma alternatives” where you are the reference point).
That shifts the job of performance marketing:
- From: win every generic query at any cost
- To: use generic queries to manufacture branded demand that survives intermediation
Three concrete plays
-
Design campaigns to create searchable memory.
In creative and landing pages, repeat distinctive brand + category hooks:- “Klaviyo, email and SMS for ecommerce brands”
- “Rippling, all-in-one HR and IT for fast-growing companies”
You want users later searching “Rippling IT onboarding” instead of “IT onboarding software.”
-
Dominate your own “brand + job to be done” queries.
Build specific pages for:- “[Brand] pricing”
- “[Brand] alternatives”
- “How [persona] uses [brand] for [outcome]”
These pages are catnip for AI overviews and agents answering “Is X right for me?”
-
Run media that doesn’t rely on click-through to work.
Podcasts, creator integrations, and social video where the primary KPI is branded search and direct traffic lift, not CTR. This is how you route around answer-engine compression.
Principle 4: Fix your measurement before AI finishes breaking it
Misreported ROAS, account chaos, and AI-driven auto-optimization all share a root cause: you’re outsourcing judgment to systems that don’t care about your P&L.
High-growth companies are already changing how they measure marketing. Not because it’s trendy, but because the old data is now lying more convincingly.
What a resilient measurement stack looks like now
You don’t need a 50-page MMM deck. You do need three layers that agree within a reasonable band:
-
Source-of-truth revenue view.
From your data warehouse or CRM:- Revenue and margin by cohort, not just by channel
- Basic media mix view: how does total paid correlate with total new revenue over time?
-
Incrementality, not just platform ROAS.
Use:- Geo or audience split tests where possible
- Holdout tests on branded search and high-intent campaigns
- Simple “on/off” experiments by channel or campaign type
Your goal: a short list of channels and tactics that you know add net-new revenue.
-
Directional, not absolute, in-platform metrics.
Accept that:- ROAS is biased but still useful for relative decisions inside a channel
- AI bidding will optimize to the goal you set, even if it’s the wrong one
So set goals that correlate with real value: qualified leads, trial activations, revenue proxies – not just “purchases” that include returns and discounts.
The job of the CMO and performance leader is to reconcile these three, then tell finance a story that matches the P&L, not the ad platforms.
Principle 5: Treat AI tools as operators, not oracles
The other pattern in the headlines: “What AI writing tools get wrong,” “AI’s trust problem,” “AI creative in 2026.” The risk is not that AI writes bad copy. The risk is that you quietly outsource your judgment to systems optimized for speed, not signal.
Practical guardrails for AI in your marketing org
-
AI drafts, humans decide.
Use AI for:- Variant generation (headlines, hooks, angles)
- Summaries and briefs (turning research into usable inputs)
- Format translation (turning a webinar into an email, a script into a blog)
But keep humans responsible for:
- Positioning and messaging hierarchy
- Offer construction
- What you are and are not willing to say to the market
-
Centralize “truth” before you scale AI outputs.
A single, maintained source for:- Positioning statements by segment
- Approved claims, proof points, and data
- Brand voice rules with examples
Then feed that into your AI stack so you get consistent speed, not random speed.
-
Instrument AI-produced assets separately.
Tag and track content and creatives produced with heavy AI assistance. Watch:- Engagement quality (time on page, scroll depth, reply quality)
- Down-funnel performance (lead quality, retention)
If AI content is cheaper but quietly worse at creating real customers, you need to see that before it poisons your funnel.
What to actually do in the next 90 days
If you run marketing, growth, or media, here’s a concrete 90-day plan to adapt to this new layer of intermediation.
1. Run an “intermediation audit”
- Audit top 50-100 queries and identify where AI overviews and agents are active.
- Document how your brand appears (if at all) in those answers.
- Flag 10-20 “must-win” intents where losing visibility would materially hurt revenue.
2. Make 5-10 pages truly “answer-engine ready”
- Pick your core category page, pricing page, and 3-5 high-intent topics.
- Rewrite them to be:
- Brutally clear on who you’re for and what you do
- Rich in structured data and FAQs
- Free of internal cannibalization (one page per intent)
- Monitor how they show up in AI overviews over 4-8 weeks.
3. Re-baseline your measurement
- Align with finance on a single source-of-truth revenue view.
- Pick one channel and one campaign type to run a simple incrementality test.
- Stop optimizing solely to platform ROAS; introduce at least one business-grounded KPI into your bidding logic.
4. Set AI guardrails in writing
- Define where AI is allowed: drafts, summaries, variants – and where it is not: claims, pricing, legal-sensitive content.
- Create a single, accessible “truth doc” for positioning and proof points.
- Tag AI-assisted assets so you can compare performance over time.
The operators who win the next five years won’t be the ones with the most AI tools. They’ll be the ones who understand that the real game is intermediation – and who rebuild their marketing, media buying, and measurement around that fact.