How should you choose a content agency for AI search?

Most agencies are still optimising for Google algorithms from 2019. Here's how to spot the ones who actually understand what ChatGPT is doing to your pipeline.

A B2B SaaS marketing director told me their content agency had just delivered 47 new blog posts. "Fantastic," I said. "How many times has your brand been cited when prospects search for solutions in your category?" Blank stare. The agency hadn't even considered it. They were too busy celebrating keyword rankings that prospects stopped caring about eighteen months ago.

This is the content agency selection crisis in miniature: brilliant execution of an obsolete playbook. Your buyers aren't Googling "best CRM software" and clicking through ten blue links anymore. They're asking ChatGPT "which CRM should a 50-person SaaS company with Salesforce integrations use?" and expecting a definitive answer with citations. According to G2's August 2025 survey of 1,000+ B2B software buyers, 87% report AI chatbots are changing how they research, with 50% now starting their buying journey in an AI chatbot instead of Google Search - a 71% jump from just four months prior. If your agency doesn't understand this shift, you're paying them to write content that AI systems will never surface when it matters most.

The delusion of "AI-optimised content"

Every agency claims they do AI optimisation now. It's bollocks, mostly.

What they actually mean: we asked ChatGPT to write some blog outlines, then our writers tidied them up. Or worse - they're still writing for Google's 2019 algorithm but slapped "AI-ready" on the proposal because their sales team read an article about it. The tell-tale sign? Ask them to show you where their client's content gets cited in AI responses for decision-stage queries. If they can't demonstrate this with specific examples, they're guessing.

Real AI extractability isn't about keyword density or semantic SEO or any of that tired Google-era apparatus. It's about whether Claude, ChatGPT, Perplexity, and Gemini can actually use your content when a qualified buyer asks a purchasing question. Forrester research shows that up to 90% of B2B organizations now use generative AI in some aspect of their purchasing process, with AI-driven tools growing at a rate of more than 40% per month. Can these systems find your product comparison pages? Your implementation guides? Your honest constraint documentation? If the agency you're evaluating doesn't track AI citation data for their clients, they're not doing the work that matters.

What agencies get catastrophically wrong about AI search

Most content agencies are structured around a mental model that died in 2023: the buyer's journey as a linear funnel where you intercept people at awareness stage, nurture them through consideration, and close them at decision. Write top-of-funnel thought leadership. Create middle-funnel guides. Maybe do some comparison pages.

Rubbish. Complete rubbish.

Modern B2B buyers compress the entire journey into a single AI conversation. They ask: "I need a solution for X, considering Y constraints, with Z integrations, budget around $50k, what should I evaluate?" The AI system either cites you in that response or you don't exist in the buyer's consideration set. There's no awareness stage. No nurturing. Research from 6Sense shows B2B buyers now spend 83% of their time researching independently, away from sales reps. You're either in the answer or you're invisible.

Traditional agencies write content for the old funnel because that's what their pricing model demands. They sell you 20 blog posts per month at $800 each. Deliverables are measured in volume. But AI-era content requires a completely different approach: deep, reference-grade material that addresses genuine decision friction. The kind of content that takes 40 hours to research and write, not four. According to Backlinko's analysis of 912 million blog posts, 93% of B2B content generates no external links - only 3% gets linked to multiple websites. Most agencies can't make the economics work on that, so they don't even try.

The agency evaluation framework that actually matters

Forget the usual agency selection theatre - portfolio reviews, case studies with suspiciously round ROI numbers, founders with impressive backgrounds. None of that tells you whether they can get your brand cited when it counts.

Here's what you should demand instead:

Proof of AI citation capability. Ask the agency: show me three times your content got cited by ChatGPT or Claude or Perplexity in response to decision-stage queries for your clients. Not "here's a blog post that ranks well." Show me the actual AI conversation where a buyer asked a purchasing question and the AI system referenced your client's content. If they can't produce this evidence, they're not doing the work.

Decision-stage keyword mapping. The agency should be able to articulate exactly which comparison queries, alternative searches, pricing questions, migration scenarios, and implementation queries matter for your category. Not broad topic clusters. Specific decision-stage searches where buyers are comparing you against named competitors or evaluating whether you solve their exact constraints. If they talk about "increasing organic traffic" instead of "getting cited when buyers evaluate solutions," wrong agency.

Content depth over content volume. Traditional agencies sell monthly retainers based on publishing frequency. Ten posts, twenty posts, whatever fits the budget. AI-era content strategy requires the opposite: fewer pieces of genuinely comprehensive material. One reference-grade comparison page that honestly addresses your product's constraints is worth more than fifty thought leadership posts about industry trends. The agency needs to understand this trade-off and price accordingly.

Constraint-aware positioning. This is the hardest filter. Most agencies will write marketing copy that makes your product sound universally brilliant. But AI systems cite content that acknowledges trade-offs and constraints honestly. "Our product is excellent for mid-market SaaS companies who need X but not ideal for enterprises that require Y-level security compliance." That kind of honest constraint framing is what earns citations. Research from Magenta Associates shows 71% of UK decision-makers say they would avoid suppliers who fail to provide clear, transparent information. If the agency recoils at writing honest limitation documentation, they don't understand how modern buyers research.

The diagnostic questions that expose agency competence

Right, let's get specific. When you're sitting across from an agency (virtually or otherwise), these questions will immediately separate the ones who understand AI search from the ones who read a blog post about it last week.

"How do you measure whether content is working in AI systems?" The weak answer: "We track engagement metrics and organic traffic." The competent answer: "We run systematic citation tracking - querying ChatGPT, Claude, Perplexity, and Gemini with the exact decision-stage questions your buyers ask, documenting when your content gets referenced, and measuring share of voice against competitors." ChatGPT alone processes 1.1 billion queries daily with 500 million weekly users, while Perplexity handled 780 million monthly queries in May 2025 - a 239% increase from August 2024. If they haven't built this measurement apparatus, they're flying blind.

"Show me a piece of content you've created that handles product constraints honestly." Watch them squirm. Most agencies have never written a single piece that says "our product isn't suitable for X use case" because their mental model is pure promotional marketing. But AI systems preferentially cite content that acknowledges limitations because it signals trustworthiness. An agency that can show you honest constraint documentation has crossed the conceptual bridge. One that can't is still living in the brochure era.

"What's your process for extracting decision friction from sales calls and support tickets?" This question tests whether they understand that the best content strategy comes from internal truth, not market research reports. Your sales team hears the same eight objections every week. Your support tickets reveal the implementation challenges that make buyers hesitate. According to McKinsey research, 57% of B2B sellers feel that the content their marketing team produces is generic and unresponsive. An agency that mines this internal data and turns it into content that addresses real friction is worth ten times more than one that writes generic "best practices" posts based on competitor analysis.

"How would you handle a situation where our product genuinely isn't the best choice for a prospect's use case?" If they look horrified, run. The right answer is: "We'd document that clearly in your comparison content, explain which alternatives would be better fits, and why - because when buyers see that intellectual honesty, they trust your recommendations for the use cases where you are the best choice." This is alien thinking to traditional agencies, but it's fundamental to earning AI citations.

The pricing model that reveals everything

Agency pricing structure tells you more about their competence than any case study ever will.

Traditional content agencies charge by deliverable volume: $X per blog post, $Y per whitepaper, $Z per case study. This model only makes sense if you're optimising for content quantity and Google rankings. It completely breaks down for AI-era content because reference-grade material takes genuine research time. You can't write a comprehensive comparison page that honestly evaluates your product against three competitors, includes real implementation data, and addresses actual decision friction in the same time it takes to bash out a thought leadership post about industry trends.

The agencies who understand AI search price differently. They charge for research depth and strategic value, not word count. A proper decision-stage content system might be 15-20 pieces of deeply researched material over six months rather than 60 generic blog posts. The economics are different. The deliverable cadence is different. The measurement is different.

Data from TechBullion shows AI search traffic converts at 4.4x the rate of traditional organic search - but you need the right content to capture it. If an agency quotes you based on "posts per month," they're still selling the old model. If they quote based on "decision-stage queries covered" or "competitive positioning documented" or "citation share improvement," they might actually know what they're doing.

The in-house versus agency question (and why you're probably asking it wrong)

Should you build an in-house content team or hire an agency? Wrong question, mostly.

The real question is: do you have anyone internally who understands decision-stage content strategy for AI systems? If yes, you can build in-house. If no, you need an agency to establish the methodology first, then potentially bring it in-house once the system is working. But most companies are asking "agency or in-house" before they even understand what good looks like in this new environment, which leads to spectacularly expensive mistakes both ways.

I've seen companies hire three in-house content people who spend six months producing beautiful thought leadership that never gets cited by AI systems because nobody told them that's not the game anymore. I've also seen companies pay agencies $15,000 per month to publish blog posts optimised for SEO tactics that stopped working in 2023. Ahrefs data shows that 28% of ChatGPT's most-cited pages have zero organic visibility in Google search - the game has fundamentally changed. Both scenarios burn cash. Both fail to move the revenue needle.

The smarter approach: start with an agency that can prove AI citation capability, use them to build your decision-stage content foundation, document the methodology as you go, then decide whether to bring it in-house once you actually understand the repeatable process. Trying to build in-house from scratch without a proven methodology is madness.

What the next eighteen months will do to agency selection

The content agency landscape is about to bifurcate savagely.

Agencies that adapt to AI search will become more valuable and more expensive because the skill set is genuinely harder than traditional content marketing. Writing reference-grade material that gets cited requires deeper research, better strategic thinking, and intellectual honesty that most marketing content has never demanded. The AI search engine market, valued at $43.6 billion in 2024, is projected to capture 62.2% of total search volume by 2030, with revenues nearing $379 billion. The agencies who can do this well will charge premium rates and be worth it.

Agencies that don't adapt will face catastrophic margin compression. Their clients will slowly realise that generic blog posts don't generate pipeline anymore. Research from Demand Gen Report shows 40% of marketers cite content development as their most pressing SEO issue - but the real problem is they're developing the wrong content. The agencies will respond by lowering prices and increasing volume, which makes the content worse, which makes it even less likely to get cited, which makes clients even less satisfied. Doom loop.

You're better off choosing now - even if it means paying more - than waiting until the market sorts itself out. Neil Cohen, CMO of cybersecurity firm Kasada, notes that site visitors from AI platforms spend up to three times more time on-page than those from traditional search engines. By the time everyone figures out which agencies are competent at AI-era content, you'll have already lost eighteen months of potential citations to competitors who moved faster.

The wrap-up (since you probably skimmed here first)

Choosing a content agency for AI search isn't about portfolio reviews or case study theatre. It's about finding an agency that can demonstrate actual AI citation capability, understands decision-stage keyword mapping, prices for content depth rather than volume, and can write constraint-aware positioning that acknowledges your product's limitations honestly.

The diagnostic questions that matter: How do you measure AI citations? Show me honest constraint documentation you've created. What's your process for mining decision friction from internal data? How do you handle situations where the client's product isn't the best fit?

If an agency can't answer these questions competently, they're still optimising for 2019 Google. And you're about to pay them to make you invisible in the searches that actually matter. With 66% of UK senior decision-makers now using AI platforms as part of the procurement process, the stakes have never been higher.

Want to stop guessing whether your content works in AI systems? Start tracking where your brand gets cited (or doesn't) when qualified buyers ask decision-stage questions. The agencies who can help you with that are rare. The ones who can't are everywhere.