The real battle isn’t “rankings” anymore. It’s recommendations.
Look at the headlines you’ve been skimming:
- “Why your brand isn’t making the AI recommendation set”
- “Why good content still loses in Google Search”
- “6 generative engine optimization benefits every marketer should know”
- “TikTok builds for the AI future, welcoming third-party agents for ads”
- “The state of agentic advertising”
- “YouTube wants to be a home, not a launchpad”
The pattern: distribution is shifting from “search and scroll” to “systems decide for you.”
Search results, feeds, “For You” pages, carousels, “Because you watched…”, “People also bought…”, AI answers, agents that buy for you-these are all recommendation systems.
If you’re still optimizing for “rankings” and “posting calendars” while ignoring how recommendation engines actually choose winners, you’re playing the wrong game.
From intent to inference: how distribution really works now
Three big shifts are colliding:
1. Search is becoming answer and assistant
Google’s ALDRIFT research, generative search experiences, and “why good content still loses” stories all point to the same thing: Google is less a directory and more an answer engine.
That means:
- You’re not trying to “rank” as a blue link; you’re trying to be cited, summarized, or recommended inside an AI-generated response.
- Entity understanding (who you are, what you’re about) matters more than exact-match keywords.
- “Good content” that isn’t machine-readable and entity-strong can still lose to mediocre content that is.
2. Feeds are now predictive, not reactive
TikTok, YouTube, Instagram Reels, Shorts, LinkedIn, even email clients-everything is tuned to predicted watch time, engagement, and satisfaction.
That’s why we see:
- Guides on “the science of attention” for short-form video.
- Data on “best time to post” and “video specs” that matter less than your ability to generate signals the model cares about.
- YouTube explicitly saying it wants to be a “home, not a launchpad”-translation: keep users in our recommendation loop.
3. Agents are quietly becoming media buyers and merchandisers
TikTok’s MCP server for AI agents, “agentic advertising,” Notion and Microsoft wiring agents into workflows, Amazon reversing API fees under pressure from tech firms that depend on programmatic access-this is the early shape of a world where:
- AI agents research, compare, and buy without your user ever “visiting” you in a traditional sense.
- Recommendation sets are built for machines first, humans second.
- Being machine-preferred becomes as important as being human-preferred.
Net result: the core marketing question is shifting from “How do I get discovered?” to “How do I become the default recommendation?”
What recommendation systems actually optimize for
Different platforms, similar math. The models vary, but the incentives rhyme. At a high level, recommendation engines care about:
- Engagement quality – not just clicks, but watch time, dwell time, saves, replies, long reads, repeat opens.
- User satisfaction – explicit (likes, ratings, “not interested”) and implicit (do they bounce? do they come back?).
- Session value – does your content keep users in a high-value loop for the platform (more time, more ads, more transactions)?
- Reliability and safety – low spam, low complaints, low policy risk, consistent performance.
- Entity clarity – can the system confidently classify who you are, what you sell, and who you serve?
Your job is to translate these into levers you can actually pull across search, social, and AI surfaces.
From SEO and “posting” to Recommendation Set Optimization (RSO)
Think of this as RSO: Recommendation Set Optimization. Not a new buzzword; a practical reframe.
The question: “What would we do differently if our only goal was to be the obvious choice for recommendation engines?”
1. Architect your brand as an entity, not just a website
We’re seeing a wave of content about schema, entities, and AI citations. The punchline: the web is being re-indexed around entities, not pages.
Action checklist:
- Own your entity graph: Ensure consistent name, category, and descriptions across your site, Google Business Profile, LinkedIn, Crunchbase, app stores, marketplaces, and major directories.
- Implement schema where it matters: Product, Organization, FAQ, HowTo, Review, Event-prioritize templates that map to real commercial intent.
- Feed the machines with context: Clear “About” and “Who we’re for” pages; explicit verticals, use cases, and segments. Ambiguity is the enemy of recommendation.
- Structured performance proof: Case studies and testimonials with measurable outcomes, industries, and roles-things an AI can safely quote.
You’re not doing schema “for rich snippets.” You’re doing it so AI systems can confidently say, “This brand is a good answer for this kind of user with this kind of problem.”
2. Design content for “stickiness per impression,” not volume
Feeds and AI surfaces don’t care how many posts you ship; they care how each impression performs.
Shift your team’s scorecard from “pieces published” to:
- Average watch time / completion rate for video.
- Scroll depth / dwell time for articles and landing pages.
- Save / share / reply rate for social and email.
- Return visits and branded search following exposure.
Practical moves:
- Short-form video: ruthlessly front-load tension (problem, promise, or pattern interrupt) in the first 1-2 seconds; cut intros; design for sound-off.
- Long-form / SEO: open with a specific outcome and who it’s for; strip generic intros; add jump links and scannable structure; answer the question fast before expanding.
- Landing pages: prioritize clarity over cleverness; make the primary action and primary proof visible without scroll; reduce cognitive load per section.
The algorithm’s “attention” is a lagging indicator of your user’s attention. Fix the latter; the former follows.
3. Build for machine trust, not just human trust
AI systems are conservative. They prefer sources that feel:
- Stable (not constantly changing domains, branding, or structure).
- Well-cited (linked from other authoritative entities).
- Low-risk (no spam signals, policy violations, or wild swings in behavior).
For CMOs and performance leaders, that means:
- Stop nuking your own history: Frequent rebrands, URL restructures, and content purges reset the trust you’ve built. If you must change, migrate carefully and preserve signals.
- Invest in “boring” hygiene: Technical SEO, site speed, uptime, clean tracking, and consistent naming conventions. These are table stakes for machine trust.
- Curate your external graph: PR, podcasts, partner listings, and thought leadership that clearly tie your brand to specific problems and categories.
4. Treat AI agents as a new performance channel
TikTok’s AI agents, Amazon’s API saga, Claude skills, and Chrome/Edge copilots all hint at the same future: a growing share of “demand” will come from agents doing the research and shortlisting.
Start acting like that’s already true:
- Write for “agent questions” explicitly: Add sections like “Who this is for,” “When this is a bad fit,” “How we compare to X/Y/Z,” “Total cost of ownership,” and “Implementation timeline” to your key pages.
- Standardize your facts: Pricing ranges, contract terms, SLAs, integrations, compliance-present them in consistent formats that are easy to extract.
- Expose clean APIs where it makes sense: Product feeds, inventory, pricing, availability, documentation. If agents can query you directly, you reduce friction to being recommended.
- Monitor AI surfaces: Periodically ask major models and assistants how they’d solve problems in your category. Where you’re missing or misrepresented, you’ve found a to-do list.
5. Align media buying with recommendation dynamics, not just auction math
Paid media is increasingly a way to seed and accelerate organic recommendation, not just to buy one-off clicks.
Rethink your playbook:
- Optimize for post-click behavior, not just CTR: Campaigns that generate strong dwell time, high completion, and low bounce feed the models better signals than cheap accidental clicks.
- Use paid to train the algo on the right audience: Start with tightly defined, high-intent cohorts; once you see strong engagement, expand lookalikes. You’re teaching the system who you’re “for.”
- Design creative for the feed’s goals: On TikTok or Reels, aim for watch time and replays; on YouTube, for session extension; in search, for fast answer and high task completion.
- Measure “halo” effects: Track branded search, direct traffic, and organic recommendation lift following paid bursts. If your reporting only sees last-click ROAS, you’ll underinvest in the campaigns that actually move you into the recommendation set.
What to change in the next 90 days
This doesn’t require a five-year roadmap. It requires a ruthless reprioritization of what your teams work on now.
For CMOs
- Reframe the mandate: Ask each channel owner, “How are we becoming the default recommendation in your environment?” Make them show how their roadmap maps to entity clarity, engagement quality, and machine trust.
- Fund the unsexy work: Approve budget for technical cleanup, schema, analytics fixes, and content refactoring. These are no longer “nice-to-have.” They’re how you exist in AI answers.
- Align narrative with data: Your “company narrative” should be specific enough that a model can summarize it in one sentence. If it sounds like it was written for an awards entry, it’s probably useless to a recommender system.
For performance marketers and media buyers
- Change your primary KPI per channel: For video, focus on completion and watch time; for search, on post-click engagement; for social, on saves/shares and profile visits.
- Run “stickiness sprints”: Take your top 10 traffic-driving assets and run structured experiments to improve dwell time and conversion. Treat every 10% gain as compounding recommendation fuel.
- Instrument for agent-era analytics: Tag and track visits and conversions from AI surfaces where possible (e.g., referrers, UTM conventions, assistant-specific URLs). You want an early read on how much business is coming from non-human journeys.
For growth and product leaders
- Make your product “explainable” to machines: Clear feature naming, public documentation, consistent plan structures, and transparent limitations. Ambiguous, clever naming is cute until an AI can’t classify you.
- Ship agent-facing surfaces: Developer docs, partner APIs, or even simple CSV/JSON exports that make it easy for third-party agents and tools to include you in their recommendation sets.
- Close the loop with CX: Feed real outcomes, NPS, and retention data back into your marketing stack. Recommendation systems reward brands that keep users happy after the click.
The platforms are telling you where they’re going: more AI, more agents, more recommendation-driven journeys. The question is whether your brand is architected to be chosen by systems that don’t care about your calendar, your campaigns, or your slogans-only about whether you’re the safest, clearest, highest-yield recommendation for a specific user in a specific moment.
Stop asking, “How do we rank?” Start asking, “Why would any system pick us?” Then build for that answer.