How Marketing Agencies Can Prepare for AI Search

By Ali Jakvani, Cofounder

Most agency P&Ls assume a content-and-links motion priced against monthly retainers, with rank tracking as the proof artifact. AI search breaks every part of that stack at once. Adding "AEO" as a feature on top of an unchanged motion is the same failure pattern that killed link-shop retainers when content marketing arrived in 2014.

What a 2026 agency needs to deliver

A defensible AI-era agency offering has four layers. None are optional. If your current offer cannot map onto these rows, you are selling a 2018 product against a 2026 market.

Layer 1: The AEO audit as productized intake

The audit is the entry point. It also fixes a structural problem most agencies have, which is that intake calls produce vague scopes and mispriced retainers. A productized AEO audit answers six questions about a client domain.

• Crawler reach. Which AI agents can actually fetch the site? GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Bingbot.
• Render parity. What does the rendered HTML look like to a non-JS bot? Is the answer present in the initial HTML?
• Structural extractability. Are H2s standalone questions or claims? Are direct answers in the first 60 words of each section?
• Schema coverage. Are Article, Organization, Person, FAQPage, BreadcrumbList present and valid across templates?
• Entity coherence. Is the brand a stable, disambiguated node? Are product names consistent across pages?
• Citation baseline. On a defined prompt set, how often does the domain appear as a citation across ChatGPT, Perplexity, Gemini, and Google AI Overviews?

Layer 2: Content engineering, not content production

Most agency content teams are structured around volume. Briefs in, drafts out, edits, publish. That model produces narrative content that is poorly engineered for extraction.

The QA step is the one most agencies skip. Before publishing, run each H2 section through an LLM with the prompt: "Answer the question X using only this passage. If you cannot, say why." If the model cannot answer cleanly, the passage is not extraction-ready and needs a rewrite.

Layer 3: Entity and schema infrastructure

This is the layer most generalist agencies do not staff for, and the layer that creates the most defensible client lock-in. The work is engineering work.

• Build a canonical entity inventory (brand, products, people, locations, services).
• Resolve naming conflicts across the site.
• Add sameAs references that connect entities to authoritative external nodes.
• Implement JSON-LD across templates, not as one-off page additions.
• Validate structured data programmatically in CI.
• Ensure render parity for AI agents that do not execute JavaScript.
• Make sitemap lastmod entries reflect actual changes.
• Build internal linking that follows entity relationships, not just topical proximity.

Layer 4: Citation monitoring as the reporting layer

Rank tracking does not justify a 2026 retainer. The reporting artifact has to show citation movement across the engines that actually carry the prompts. A useful monitoring stack covers four signals.

This requires probing each target engine on a defined prompt panel at a regular cadence. Internal tooling can do it, or AI visibility scoring systems can maintain the prompt panels and historical deltas for you. Either way, the dashboard is the artifact the client looks at every month.

Prompt panel design is the new keyword research

Build panels that include direct branded prompts ("what is brand X"), comparison prompts ("brand vs competitor"), category prompts ("best category for persona"), problem prompts ("how do I X"), and long-tail prompts that map to commercial intent. 50 to 200 prompts is a reasonable starting size. Track them weekly.

Agency economics

The mix that holds together: audit pulls the client in, engineering and infrastructure produce the lift, monitoring justifies the ongoing fee. Trying to price this as a flat content retainer misses the engineering value and undercharges for the durable infrastructure work.

What dies inside the agency

• Pure rank trackers. A weekly position report does not justify a retainer.
• Generalist content writers without retrieval discipline. 2,000-word narratives without passage engineering do not move citation rate.
• Link building as a standalone offering. Backlinks remain useful but cannot carry a retainer.
• One-off blog package selling. Per-post pricing does not align with the cluster-and-entity nature of AEO.

What replaces them: AEO analysts, content engineers, schema and entity engineers, and visibility analysts who own the citation dashboard.

A 90-day agency upgrade plan

Days 1 to 30: Internal readiness

• Pick three reference clients across different verticals to pilot the new stack.
• Define a target engine set (typically ChatGPT, Perplexity, Gemini, Google AI Overviews).
• Build the prompt panel template (50 prompts per client).
• Choose the citation monitoring stack (build vs buy).
• Train one writer and one technical lead on passage engineering and schema implementation.
• Draft the productized AEO audit deliverable template.

Days 31 to 60: Client pilots

• Run the AEO audit on each pilot client.
• Implement the entity and schema infrastructure on one template per client.
• Rewrite five passages per client using extraction-ready structure.
• Stand up the citation monitoring dashboard.

Days 61 to 90: Productization

• Codify pricing for audit, engineering, infrastructure, and monitoring.
• Build the sales narrative around the citation share chart.
• Update the agency website to demonstrate AEO competence (eat your own cooking).
• Run the first monthly citation review with each pilot client.

Frequently asked questions

Do agencies need to drop classical SEO?

No. Classical SEO is a subset of AEO. Most retrieval pipelines pull from the same web Google indexes. The error is treating ranking as the deliverable rather than as one input to citation probability.

How big a team do you need to deliver AEO?

A minimum viable AEO pod is one analyst, one content engineer, one schema/entity engineer, and shared monitoring infrastructure. Smaller agencies can have one person wear two hats, but all four functions must exist.

How do you prove AEO ROI to a CMO?

Tie citation share movement to pipeline. The flow is citation share → AI-influenced traffic and brand mentions → downstream pipeline. Closed-loop attribution is hard, but citation share is measurable, longitudinal, and competitive, which is what a CMO needs to fund the line item.

Should agencies build or buy citation monitoring?

Build if you have engineering capacity and want a long-term moat. Buy if you need to ship in 90 days. Most agencies start with buy and migrate to a hybrid stack later.

What is the most common implementation mistake?

Adding FAQ schema and calling it AEO. Schema is necessary but not sufficient. Without passage engineering, entity coherence, and citation monitoring, FAQ schema is a single tactic, not a service.