Beyond AEO: The Agentic Recommendation Journey

Tom Williams, CEO

If you’re only just starting to think seriously about AEO and GEO, you may be focused on solving yesterday’s problem.

For the last twenty years, digital discoverability was relatively straightforward. You wanted people to find you, so you made sure search engines could find you first. SEO became an industry, companies learned to structure websites around search intent, and Google became the gateway between a question and an answer.

Generative AI has changed that. Increasingly, people no longer search, click, compare, and read their way to an answer. They ask ChatGPT, Claude, Gemini, Perplexity, or another AI system to do some of that work for them.

That shift has created the rapidly growing fields of AEO and GEO: making sure your company and content can be understood, retrieved, cited, and surfaced by AI. But that challenge is already moving again.

I recently wrote about the new developer adoption gap in the Catchy H2 2026 Insights Report. I explore how AI has accelerated the supply of software without equivalently accelerating our ability to understand, validate, maintain, and adopt what gets built. The bottleneck has moved downstream.

Something similar is now happening to AI discovery. Getting found by AI has rapidly become table stakes. The more interesting question is what happens after the model finds you.

Because increasingly, AI is not simply retrieving a piece of information. It can compare sources, reconcile evidence, incorporate third-party perspectives, and synthesize a much wider picture of your product.

The challenge is shifting from visibility to validation. And that means you can win the citation and still lose the recommendation.

Three Era's of Discovery

Over the last twenty years, the challenge of digital discoverability has evolved through three distinct phases.

The SEO era (c. 2005-2022)

Can I be found?

The first job was accessibility. Could a search engine crawl your site, understand what a page is about, and decide that it is relevant enough to appear when somebody searches? The unit of optimization was largely the page, and success meant visibility.

The AEO & GEO era (c. 2023-2026)

Can I be surfaced?

Generative AI changed the interface. The question became whether AI systems could retrieve, understand, cite, and surface your content when somebody asks an agent a question. That created a new focus on machine-readable content, structured information, authoritative documentation, and citability. Developers increasingly start with AI, allowing it to synthesize sources and guide them toward the information they need.

But this is where much of the current conversation stops too early.

The synthesis era (2026 onwards)

Can I be validated strongly enough to be recommended?

Once an AI system can find you, it does not have to take your own content at face value. It can read your documentation and inspect your repositories. It can see what developers say about you, find third-party tutorials and reviews, look at integrations and ecosystem activity, and compare your claims with the wider body of evidence around your product.

That changes the problem. The question is no longer simply whether AI can find or surface you. It is whether the evidence it finds coheres strongly enough for it to recommend you. The question shifts from “Can the model find me?” to

When the model synthesizes the available evidence about me, what story emerges?

The Agentic Recommendation Journey

This creates a new journey for companies to consider. It has four stages, each with a higher bar than the last.

The first two stages are familiar. In Found, the agent discovers and interprets information about your product from the surfaces you control: your website, documentation, repositories, and structured content. In Surfaced, it decides you're relevant enough to include in an answer or consideration set. This is where much of today's AEO and GEO work is focused, and where much of it stops.

Then things get more interesting.

In Validated, the agent starts corroborating the claims it has found, drawing on GitHub, Reddit, Stack Overflow, developer communities, technical creators, reviews, benchmarks, and integrations. This is the stage many current AEO strategies underweight: it depends less on what you publish and more on whether the wider evidence supports it.

In Recommended, the agent synthesizes everything into a judgment about which product, platform, API, or service best fits the task. At this point there is no single surface to optimize. The recommendation is the output of the whole system.

AEO and GEO help you get surfaced. Recommendation requires validation.

AI is becoming the synthesis layer

This is the important conceptual shift.

AI is often drawn as another touchpoint in the developer journey, sitting alongside search, documentation, community, GitHub, events, or social media. I increasingly think that is the wrong model.

AI is becoming the synthesis layer across the journey.

A developer previously had to do much of this synthesis manually. They might encounter a product through search, read its documentation, check whether the repository looked healthy, search Reddit for complaints, ask a colleague, watch somebody build with it, inspect its integrations, and eventually decide whether the accumulated evidence was convincing. AI can increasingly compress that process.

That changes what constitutes optimization. The first wave of AEO understandably focused on content: make your documentation easy to retrieve, structure pages clearly, answer questions explicitly, improve the likelihood that an AI system cites you. All of that remains useful. But well-structured documentation is only one signal.

If your docs say integration takes five minutes and developer conversations suggest otherwise, an agent can potentially see both. If your marketing claims a thriving developer ecosystem but your repositories are dormant and third-party implementation content is scarce, those signals do not reinforce one another. If your website claims broad compatibility but the ecosystem evidence is thin, that gap becomes increasingly legible.

The agent does not have to accept the strongest version of the story you publish. It can synthesize the story the evidence supports.

What the agent validates

This is where the wider developer program starts to matter.

I think there are six broad families of signals that increasingly contribute to agentic validation:

Documentation & Technical Content

Can the agent understand what the product does, how it works, what it supports, and how somebody would implement it?

Developer Experience & Product

Does the actual experience support the promise? Is the API coherent? Does onboarding work? Are the SDKs maintained? Can developers successfully get to value?

Community & Developer Sentiment

What are practitioners saying when the brand is not controlling the conversation? What problems recur? What do developers recommend to one another?

Third-Party Proof

Are credible people demonstrating the product independently? Are there tutorials, benchmarks, reviews, case studies, or other artifacts that substantiate its claims?

Ecosystem & Integrations

Is there evidence that the product works within real developer stacks? Which integrations exist? Which platforms support it? Where is it actually being used?

Brand & Market Signals

Does the wider market clearly understand what the company does? Is it present in the relevant conversations, categories, comparisons, and communities?

This matters because third-party evidence already appears to play a disproportionate role in AI visibility, and the external research points the same way. Ahrefs' analysis of 75,000 brands found branded web mentions correlated roughly three times more strongly with AI Overview visibility than backlinks (0.664 versus 0.218), with the strongest signals all sitting off the brand's own site. Community platforms are just as prominent on the citation side: a study of 30 million cited sources found Reddit the single most-cited source across ChatGPT, Google AI Mode, Gemini, Perplexity, and AI Overviews. Among Perplexity's most-cited sources, Reddit alone accounts for roughly 47%. These findings are correlational rather than causal, but the direction is consistent; the next logical step is that these signals will increasingly be weighed together.

Peaks help you get surfaced. Coherence helps you get recommended.

Imagine two companies.

Company A has exceptional documentation and has invested aggressively in AEO. Its technical pages are beautifully structured, easily retrieved, and frequently cited. But its developer community is weak, its third-party tutorials are scarce, its integrations are inconsistent, GitHub activity is patchy. And the actual product experience creates friction.

Company B is slightly less impressive on documentation alone. But its docs are good, developers speak positively about it, its repositories are healthy, technical creators can demonstrate it successfully, its integrations are easy to verify, and its product behaves as promised.

Which company should an agent recommend? Increasingly, I think it will be Company B. That is the more important implication of the synthesis era: agentic recommendation rewards coherence, not just peaks.

This does not mean every element of a developer program has to be perfect. Nor will every signal carry equal weight for every query. But it does mean a company can no longer assume that excellence in one highly optimized channel will compensate indefinitely for weakness elsewhere. Your whole outward-facing developer footprint is becoming evidence.

And the more capable AI systems become at synthesizing that evidence, the harder inconsistencies will be to hide.

The implications for developer marketing

This expands the remit of AI optimization considerably.

The work cannot sit solely with an SEO team. It cannot be solved purely by rewriting documentation for AI retrieval. And it cannot end when your brand starts appearing in ChatGPT answers.

Developer marketing increasingly has to think about the entire system that surrounds a product: documentation, DX, content, community, ecosystem, advocacy, product proof, and discoverability. Those functions have often been managed independently. From the perspective of an AI system synthesizing evidence, they are increasingly part of the same story.

That creates a new strategic question for developer programs: how coherent is the evidence surrounding our product?

Because a strong developer brand is no longer simply what you say about yourself. It is what your documentation explains, what your product demonstrates, what developers experience, what communities discuss, what third parties can prove, and what your ecosystem makes visible.

That is also why I think this shift creates a much bigger opportunity than simply “doing GEO better.” The companies that act early will examine the entire recommendation surface, identify contradictions and weak signals, and deliberately strengthen the evidence agents will encounter. The companies that do not may discover that they are perfectly visible to AI, but that AI just does not recommend them.

Being findable is no longer enough

For years, the question was whether search engines could find you. Then it became whether AI systems could surface you. The next question is harder.

An agent may increasingly search, compare, interpret, cross-check, and synthesize the evidence before a developer ever visits your site. So your documentation needs to agree with your product. Your product needs to agree with your community. Your claims need to agree with the evidence. And the wider signals surrounding your brand need to reinforce a story an AI system can confidently act on.

The challenge is no longer simply being found by AI. It is being validated strongly enough to be recommended.

Having found everything, does AI have enough evidence to recommend you?

That is the question developer marketing teams should be preparing for now. If you are interested in learning more about how you can move your products beyond being found to being recommended then we'd love to talk.