AI Broke the Old Media Plan: How to Build an “Agentic” Performance Engine Instead

The pattern in all the noise: AI isn’t a channel, it’s a buyer

Scan those headlines and one thing jumps out: everyone is busy tweaking tactics (TikTok sale, Instagram tools, Pinterest stats, negative keywords, AI keyword prompts) while the ground truth quietly shifted.

The real story is this: AI is becoming an agentic buyer and gatekeeper, not just a content or ad generator.

• Google’s AI search is rewriting how links and context appear.
• Agentic AI is forecast to drive $1T in “agentic commerce” by 2030.
• AI assistants (Claude, ChatGPT, Perplexity, etc.) are becoming the first stop for “what should I buy?”
• Platforms are quietly deciding which AI bots can crawl your site.
• CPAs and CPCs are climbing as “AI Max” and similar products optimize for the platform’s economics, not yours.

For CMOs, performance marketers, and media buyers, this means the old mental model – “I buy impressions, optimize for clicks, harvest intent” – is aging out. You’re increasingly selling to, and through, machines that act like consumers.

This piece is about how to operate in that world: how to build an “agentic” performance engine that assumes AI systems are the new intermediaries between your brand and your buyer.

From channels to agents: what actually changed

Historically, your media plan revolved around:

• Channels: search, social, display, email, affiliates.
• Formats: video, feed ads, stories, PLAs, native.
• Human behavior: queries, scrolls, opens, clicks.

Now, a growing share of that behavior is being pre-processed by AI systems:

• Search is turning into answers. AI overviews, summaries, and chat interfaces are compressing 10 blue links into one machine-curated response.
• Assistants are becoming shoppers. “Find me the best running shoes under $150 for flat feet” is no longer a query; it’s a task handed to an agent that will shortlist, compare, and sometimes transact.
• Media buying is increasingly algorithmic. Products like Google’s AI Max, Advantage+ Shopping, and Criteo’s AI-driven retargeting are optimizing bids and placements with minimal human input.
• PR and content are now machine-consumed first. AI crawlers, summarizers, and RAG systems ingest your content long before a human sees it.

The uncomfortable implication: you’re no longer just persuading humans; you’re persuading systems that interpret, rank, and route human intent.

The new performance stack: design for AI intermediaries

If AI is the new gatekeeper, then your performance engine needs to do three things well:

1. Be machine-readable (so AI can understand and represent you accurately).
2. Be machine-appealing (so AI systems prefer you when they construct answers and recommendations).
3. Be machine-resilient (so your economics survive when black-box optimizers chase their own objectives).

1. Make your brand machine-readable

Most teams are still optimizing for humans who read pages. You now need to optimize for models that ingest and vectorize content.

Priority moves:

• Fix technical discoverability for AI crawlers.
  • Audit robots.txt and security plugins on your CMS (especially managed WordPress) to confirm major AI bots and search crawlers are not blocked unintentionally.
  • Segment: allow indexing of evergreen, product, and help content; restrict only true proprietary or compliance-sensitive assets.
• Structure your content like a knowledge base, not a brochure.
  • Clear, explicit entities: product names, use cases, industries, personas.
  • Declarative statements that models can quote: “[Product] is a [category] that helps [persona] do [job] by [mechanism].”
  • Consistent schema markup (Product, FAQ, HowTo, Organization) to reinforce meaning.
• Clean up cannibalization and duplication.
  • Multiple similar pages dilute your topical authority in both search and AI summaries.
  • Consolidate into fewer, stronger canonical pages per topic; redirect the rest.
• Instrument your site for AI-era analytics.
  • Track traffic from AI search features separately where possible (e.g., AI overview referrers, chat-based browsers).
  • Monitor which pages are being cited or excerpted by AI tools (where you can see it) and treat those as “upstream” assets.

The goal: when an AI system builds a vector representation of “what your brand is and does,” it’s accurate, differentiated, and anchored in a small set of strong, consistent sources.

2. Make your offer machine-appealing

AI agents optimize for outcomes: relevance, clarity, safety, user satisfaction. They are conservative by default. Your job is to make your brand the safest, clearest answer.

Think in terms of “agent incentives”:

• Clarity beats cleverness.
  • Ambiguous positioning (“platform for growth”, “solutions for innovation”) is noise to a model.
  • Use concrete categories and jobs-to-be-done: “B2B email deliverability platform for SaaS companies” is legible to both humans and models.
• Proof beats puffery.
  • Case studies with explicit metrics (“37% lift in inquiries”, “26% revenue growth from X channel”) give LLMs quotable facts.
  • Third-party mentions, PR, and reviews are now inputs to AI rankings, not just social proof for humans.
• Completeness beats minimalism.
  • AI agents prefer answers that cover constraints, trade-offs, and edge cases.
  • Build content that explicitly addresses “who this is for / not for,” pricing ranges, implementation effort, and integrations.
• Consistency beats channel-specific copy.
  • Ensure your positioning, category, and core claims match across site, LinkedIn, PR, app stores, and partner listings.
  • Inconsistent descriptions confuse entity resolution and weaken your presence in AI-generated comparisons.

If you were an AI agent trying not to embarrass yourself, which vendors would you recommend? The ones that are clear, well-documented, low-risk, and well-reviewed. Engineer your footprint to look like that.

3. Make your media buying machine-resilient

Platforms are pushing more automation: AI Max, Advantage+, “smart” bidding, dynamic creative, auto-applied recommendations. These are not neutral. They optimize for:

• Platform revenue (more spend, higher CPCs/CPMs).
• Short-term engagement metrics.
• Attribution models that favor their own inventory.

Your job is to keep control of constraints and signals.

Operational moves:

• Get ruthless with negative keywords and exclusions.
  • In an AI-optimized auction, sloppy negatives are now an expensive mistake, not a minor leak.
  • Build and maintain a living negative keyword taxonomy: brand terms to protect, competitor terms to avoid/target selectively, low-intent junk, and AI-chat queries that never convert.
  • Apply similar discipline to placement and audience exclusions in social and display.
• Feed the machine better conversion signals.
  • Move beyond “any form fill” as a conversion. Train your systems on qualified events: SQLs, pipeline, revenue.
  • Use offline conversion imports and CRM integrations so the bidding AI optimizes for business value, not vanity leads.
• Exploit high-intent call, message, and lead assets – but with guardrails.
  • Call and message extensions can be gold, but only if you track call quality and duration, not just volume.
  • Lead forms should be scored and synced to CRM within minutes; slow follow-up kills the economics that the algorithm is trying to optimize.
• Run “AI vs. human” budget splits.
  • Deliberately split campaigns into: AI-heavy (broad, automated) and human-curated (tight match types, manual bids, handcrafted audiences).
  • Measure marginal ROAS and CPA by cohort. Don’t let AI Max quietly eat the whole budget just because it’s easy.
• Protect brand and margin during CPC inflation.
  • Ringfence brand search and core product terms with strict bid caps and separate budgets.
  • Shift some spend to owned demand capture (email, community, direct) where you control the economics.

Automation is a tool, not a strategy. Your strategy is the constraints you set, the signals you send, and the economics you’re willing to tolerate.

Content and PR in an AI-first discovery world

One underpriced lever right now: treating PR and content as training data for the systems that will recommend you.

Instead of chasing vanity coverage, aim for:

• Topically aligned, high-signal mentions.
  • Get cited in pieces that answer the exact questions your buyers ask assistants: “best X for Y,” “how to do Z,” “alternatives to A.”
  • Ensure your brand is mentioned with your category and your core differentiators in the same paragraph.
• Structured, quotable narratives.
  • Provide journalists and partners with simple, factual soundbites: metrics, rankings, specific outcomes.
  • These become the sentences LLMs lift into their responses.
• Owned “canonical” explainers.
  • Publish definitive guides on the problems you solve, written to be both human-friendly and model-friendly (clear headings, FAQs, explicit definitions).
  • Update them regularly; stale content is less likely to be surfaced by AI systems tuned for recency.

Think of every high-authority page that mentions you as a training example. Your job is to increase the number and quality of those examples.

Org design: who owns “selling to machines”?

The awkward truth: most org charts are still built for a pre-agentic world. You have:

• SEO in one corner.
• Paid search and social in another.
• Content and PR somewhere else.
• Data and analytics as a service function.

Meanwhile, AI systems don’t care about your org chart. They see one messy graph of entities, content, links, and behavior.

Practical moves you can make this quarter:

• Create an “AI discovery” working group.
  • Include SEO, paid, content, PR, and analytics.
  • Give them one mandate: improve how AI systems see, describe, and route to your brand.
• Define a shared ontology of your business.
  • Agree on canonical names for products, features, personas, use cases, and categories.
  • Enforce this in all external-facing assets: site, ads, decks, PR, partner listings.
• Appoint an “agentic media” owner.
  • Someone senior in performance who is responsible for how you use platform automation, what signals you feed it, and where you draw the line.
  • Measure them on incremental profit, not just spend efficiency.
• Invest in internal AI tooling for operators, not just creative.
  • Use tools like Claude Code / Cowork and similar to build internal scripts for log analysis, query mining, negative keyword management, and CRO testing.
  • Free your best operators from spreadsheet drudgery so they can think about strategy and constraints.

What to do in the next 90 days

To make this concrete, here’s a 90-day roadmap you can hand to your team:

1. Audit your machine visibility.
   • Check robots.txt, firewalls, and plugins for blocked AI/search crawlers.
   • Inventory your top 100 URLs by traffic and conversions; ensure they’re crawlable, canonical, and internally linked.
2. Standardize your positioning language.
   • Write a one-page “model brief” that defines your category, personas, use cases, and 3-5 proof points.
   • Push this into your site copy, LinkedIn, PR boilerplate, and partner listings.
3. Refactor one core funnel for AI-era performance.
   • Pick your highest-value product or segment.
   • Clean up cannibalized content, build a canonical explainer, and align ad copy, landing pages, and sales scripts around the same language.
4. Rebuild your negative keyword and exclusion strategy.
   • Mine search terms and placements from the last 6-12 months.
   • Create and maintain structured lists by intent and quality; implement across campaigns.
5. Upgrade conversion signals in at least one major ad platform.
   • Hook up offline conversions or CRM events.
   • Shift optimization from “lead” to “qualified lead” or “opportunity” where volume allows.
6. Brief PR/content with an AI lens.
   • For your next 3-5 pieces of coverage or big content assets, explicitly design for quotable facts and clear categorization.

The teams that win the next cycle won’t be the ones with the most AI-generated ads. They’ll be the ones that understand they’re now selling to – and through – machines, and quietly rebuild their performance engine around that fact.