In-house SEO means a team that sits inside the company, on payroll, with access to the product roadmap and the analytics warehouse. Most people assume that makes it the expensive choice in the in-house vs agency SEO AI debate — and until about eighteen months ago, they were right.
They aren't anymore. The thing AI changed isn't the quality of SEO work. It changed who can produce it, and how fast, and for how much. That single shift quietly rearranged the math behind every staffing decision a marketing leader is about to make.
The old cost gap was the whole argument
For most of SEO's history, the agency case wrote itself. A mid-market company couldn't justify a full stack of specialists. So you rented one. Dan Wiggins' 2025 breakdown of a typical UK in-house function — two SEO specialists at £70,000–£90,000, a content writer at £25,000–£35,000, tooling at £8,000, training at £5,000–£10,000, plus £15,000–£18,000 in National Insurance and pensions — lands around £120,000–£140,000 a year. Against an agency retainer in the £5,000-a-month range, the agency wins on paper. It always did.
That was the whole argument. Not strategy. Not control. Cost.
AI didn't make in-house teams cheaper in headcount terms. It made each head do the work of three. A content strategist with a decent GEO workflow ships what used to take a pod. A technical engineer who knows how to prompt against log files audits a site in an afternoon. The retainer math survives. The productivity math doesn't.
⚖️ In-House vs Agency SEO: Cost & Productivity Math
What agencies are actually selling now
Strip the marketing language away and most AI SEO agencies are selling two things: access to tools, and access to people who already know how to use them. Bar Maimon's Forbes Council piece on in-house AI SEO teams vs. agencies makes this explicit — agencies offer "R&D investment, multi-client experience and more affordable access to advanced tools." That's an honest list. It's also a depreciating one.
Tools are commoditizing. The GEO Toolkit category Lumar sits in — built to help businesses get understood and cited by AI — is no longer the moat it was in 2023. Workflow knowledge is commoditizing too, because the workflows are documented everywhere and the models themselves will teach you. Multi-client experience still matters, but it cuts both ways: an agency learns from your data and then carries that learning to the next client in your category.
What an agency cannot sell you is the thing that compounds. Knowledge of your product. Of your customer's actual language. Of why the legal team killed the comparison page in 2022. That sits inside the building or it doesn't exist.
In-house vs agency SEO AI, reframed
The question isn't whether AI lets agencies do more for less. It does. It's whether AI lets in-house teams do more for less faster than agencies can pass those savings on. The answer, almost everywhere, is yes — because the in-house team owns the data the AI needs.
Will Rice, writing for Lumar's enterprise blog, observed that most SEO teams at enterprise companies still struggle to make the most of AI. He's right, and it's worth asking why. It isn't because the tools are hard. It's because the highest-leverage AI work in SEO requires plumbing — feeds from the CMS, the analytics warehouse, the product catalog, the support tickets, the CRM. Agencies get sanitized exports. In-house teams get the live pipe.
The category-level shift makes this sharper. As the Grizzle team noted in their 2026 content-services roundup, 60% of Google searches now end in zero clicks. Visibility is moving inside the answer itself — inside ChatGPT, Gemini, Perplexity. To be cited there, a model needs to find your entities, your claims, your data. Agencies can optimize copy. They can't restructure your product taxonomy on a Tuesday.
The new in-house stack does not look like the old one
Maimon's list of in-house roles — AI SEO manager, AI SEO specialist, content strategist, technical engineer, link-building specialist — is the legacy org chart with "AI" stapled to two of the titles. That's not the team you need now. The team you need now is smaller and weirder.
You need someone who understands retrieval. You need someone who can write prose a language model will quote verbatim. You need someone who treats the site as a structured data asset, not a collection of pages. The link-building specialist is mostly gone. The content writer who can't direct a model is mostly gone too. In their place: fewer people, better paid, doing work that used to require a pod.
The economics flip here. The £120,000–£140,000 figure assumed five seats. Three seats, with the right tooling, will outproduce that team. Now compare to the £5,000-a-month agency retainer — £60,000 annually for someone whose incentive is to keep the retainer, not to make themselves unnecessary.
What it takes to actually build one
This is the section worth printing out. Most companies fail at in-house SEO not because the strategy is wrong but because the first ninety days are improvised. The phases below assume you already have a product, traffic worth defending, and a leader who can hire.
Weeks 1-3: Data audit. Before anyone writes anything, map what the team can actually see. Analytics access, search console history, log files, CRM exports, product catalog structure, internal search queries. This is the input layer for every AI workflow that comes next, and if it's broken, nothing downstream will work. Most in-house teams skip this and then wonder why their AI content reads generic.
Weeks 4-8: Entity and authority mapping. Decide what your company is supposed to be the answer to. Not keywords — entities, claims, and the relationships between them. This is the work that determines whether ChatGPT cites you or your competitor when someone asks the question your sales team hears every week. It is also the work an agency physically cannot do better than you, because the source material is institutional.
Weeks 9-16: Production system. Build the workflow that turns subject-matter expertise into published, structured, model-readable output. One technical lead, one editor who can direct models, one strategist who owns the entity map. The output target is not volume. It is citation-worthy density.
Months 5-9: Distribution and feedback loops. Wire the analytics back into the production system so the team can see what AI surfaces are citing and what they're not. Most of the compounding happens here, and most of it is invisible if you haven't built the measurement layer. This is also where the team starts to look obviously cheaper than any agency you could have hired.
Month 10 onward: Compounding. The institutional knowledge accrues. The model-prompts library gets sharper. The entity map gets denser. Output per person climbs. The agency you didn't hire would be quoting you a renewal around now.
🗓️ Building an In-House AI SEO Team: First 10 Months
Map analytics access, search console history, log files, CRM exports, product catalog, and internal search queries — the input layer for all AI workflows.
Define what your company should be the answer to: entities, claims, and relationships that determine AI citation over competitors.
Build the workflow converting subject-matter expertise into structured, model-readable output. Focus on citation-worthy density, not volume.
Wire analytics back into production to track AI surface citations. Build measurement layer where compounding becomes visible and cost advantage over agencies becomes clear.
Institutional knowledge accrues, prompt libraries sharpen, entity map densifies, and output per person climbs — while an agency would be quoting renewal.
The hybrid model is a transition, not a destination
A lot of writing on this topic — including the September 2025 piece from the Why Companies Prefer AI SEO Agencies Over In-House Teams crowd, which updated again in June 2026 — lands on a hybrid recommendation. Keep some in-house, rent some from an agency. It's a safe answer. It's also usually wrong as a long-term posture.
Hybrid works during the transition, when the in-house team is still being built and an agency fills the gap. It stops working the moment the in-house team is competent, because at that point the agency is being paid to duplicate work the team can now do faster and with better context. The companies that have actually mastered organic — Airbnb, Canva, Zapier, Zillow, Hubspot all get cited in the literature for this — didn't get there by retainer. They got there by treating SEO as a core capability and staffing it accordingly, over years.
What's changed in 2026 is that you no longer need their headcount to do it. You need their patience.
The bet that compounds
The honest case against in-house has always been that it's slow to start and risky if the wrong hire makes the first three calls. That's still true. AI didn't fix hiring.
What AI did is shift the ceiling. The output a small embedded team can now produce — when it has the data, the entity map, and a leader who understands retrieval — is not in the same category as what the same team could produce two years ago. The agency ceiling moved up too. But the agency starts every engagement at zero on your context and has to be paid to climb. The in-house team climbs once and stays up there.
That's the asymmetry. Agencies rent expertise. In-house teams accrue it. AI just made the accrual rate fast enough that the math finally works.
Sources
- In-House AI SEO Teams Vs. Agencies: Which Is Best For ...
- Why Companies Prefer AI SEO Agencies Over In-House ...
- AI Use Cases for Enterprise SEO & Digital Marketing Teams
FAQ
Does AI actually make in-house SEO teams cheaper than agencies?
Not in headcount terms. AI made each head do the work of three. A content strategist with a decent GEO workflow ships what used to take a pod. The retainer math survives. The productivity math doesn't. Three seats with the right tooling will outproduce the old five-seat team — and outproduce a £60,000 agency retainer.
What are AI SEO agencies actually selling now?
Two things: access to tools, and access to people who already know how to use them. That's an honest list, and a depreciating one. Tools are commoditizing. Workflows are documented everywhere and the models will teach you. Multi-client experience cuts both ways — an agency learns from your data and carries it to your competitor.
Why can't an agency replicate what an in-house SEO team does?
Because they get sanitized exports while in-house teams get the live pipe — the CMS, analytics warehouse, product catalog, support tickets, CRM. They also can't sell you knowledge of your product, your customer's actual language, or why legal killed the comparison page in 2022. That sits inside the building or it doesn't exist.
Isn't a hybrid in-house plus agency model the safest answer?
It's safe and usually wrong as a long-term posture. Hybrid works during the transition, while the in-house team is being built. It stops working the moment that team is competent, because the agency is then being paid to duplicate work the team can do faster and with better context. Hybrid is a transition, not a destination.
What does the new in-house SEO team actually look like?
Smaller and weirder than the legacy org chart with "AI" stapled to two titles. You need someone who understands retrieval, someone who can write prose a language model will quote verbatim, and someone who treats the site as a structured data asset. The link-building specialist is mostly gone. Fewer people, better paid, doing pod-level work.
How long before an in-house AI SEO team pays off?
The first ninety days are setup — data audit, entity and authority mapping, production system. By months 5-9 you wire in feedback loops and the team starts to look obviously cheaper than any agency. Month 10 onward is where compounding hits. The agency you didn't hire would be quoting you a renewal around then.