Updated September 17, 2026: This article has been repositioned around the durable operating decision behind AI SEO: what a company should own internally, what can be outsourced, and how to compare cost, control, specialist capacity, and measurement without treating AEO or GEO as a separate Google ranking system.
The AI label does not change the core make-or-buy problem. A company still has to decide who owns product truth, search priorities, technical access, editorial approval, measurement, and the knowledge that accumulates after each change. An agency can provide specialist capacity and an outside view. An internal team can retain context and move more closely with the business. Neither advantage is automatic.
Google’s current guidance makes this distinction more useful, not less. For Google Search, optimisation for generative AI features is still SEO. Google also advises site owners to evaluate third-party AEO, GEO, and AI SEO services against official Search guidance rather than assuming a new label represents a new ranking system. The practical decision is therefore not who has the most AI terminology. It is which operating model can make good search decisions repeatedly and remain accountable for them.
The AI Label Does Not Change the Ownership Problem
An AI SEO programme may involve unfamiliar tools, but most of the responsibilities are familiar: deciding which pages deserve investment, maintaining crawlable and useful content, coordinating technical changes, checking factual accuracy, measuring search performance, and connecting that work to business outcomes.
What has changed is the number of places where a team can lose clarity. Google Search now has dedicated generative AI performance reporting. Third-party platforms offer their own AI visibility scores. Content teams can produce and revise material faster with generative tools. Agencies are adding AEO and GEO services. If nobody owns the definitions behind those activities, the reporting can become more sophisticated while the decision process becomes weaker.
I would keep four responsibilities clearly owned inside the company even when most execution is outsourced:
- Business and product truth: what the company actually sells, where it operates, what claims are accurate, and which limitations matter.
- Priority decisions: which markets, page types, products, services, and search needs deserve attention first.
- Approval and risk: who can approve technical changes, public claims, content removal, redirects, tracking changes, and access to internal systems.
- Measurement definitions: what counts as a useful outcome, which data source is authoritative, and how search visibility connects to leads, sales, subscriptions, or another business result.
An agency can advise on all four. It should not become the only place where that knowledge exists.
What an Agency Can Buy You That Hiring May Not
The strongest case for an agency is usually not lower cost. It is access to capability before the organisation can justify building every capability internally. A company may need a technical audit, international SEO review, content-system redesign, migration plan, structured data work, or AI visibility measurement without needing a full-time specialist for each area.
External specialists can also provide diagnostic distance. Internal teams naturally inherit the site’s existing assumptions, naming conventions, reporting habits, and political constraints. An outside reviewer may notice that two content programmes compete with each other, that a reporting metric is being treated as more precise than it is, or that a technical problem is consuming attention while a more important ownership problem remains unresolved.
The trade-off is context acquisition. An agency does not automatically know which product statements are sensitive, why one market requires a different approval path, which customer language is commercially important, or which CMS limitation has already been tested and rejected. Every engagement therefore carries a coordination cost even when the invoice is predictable.
This is why the useful agency question is not simply, ‘Can you do AI SEO?’ The better question is: what decisions will you make, what evidence will you use, what access will you need, and what knowledge will remain with our team when the engagement ends?
Compare the Full Cost, Not Salary Against Retainer
A salary-versus-retainer comparison leaves out too much. In-house and agency models move costs into different places. The comparison should include the work required to operate the model, not only the line item used to pay for it.
| Cost or constraint | In-house model | Agency model |
|---|---|---|
| Core labour | Recruiting, salary, benefits, onboarding, management, and retention. | Retainer, project fees, or advisory fees, plus internal time needed to manage the relationship. |
| Specialist breadth | Additional hires, contractors, or training may be needed for technical, editorial, data, localisation, or migration work. | Specialists may already be available across the agency, but their time and depth of involvement can vary by contract. |
| Tools and data | The company usually carries its own tool stack, licences, integrations, and maintenance. | Some tools may be included, but the company still needs access to first-party data and should understand how external scores are produced. |
| Business context | Context accumulates naturally if the team is stable and close to product, sales, support, and engineering. | Context has to be transferred through briefs, meetings, documentation, and review. |
| Implementation | Internal teams may coordinate changes more directly, but can be constrained by competing priorities. | Recommendations may move quickly, but implementation can stall if the agency cannot change the CMS, templates, analytics, or engineering backlog itself. |
| Knowledge retention | Methods and decisions can remain institutional knowledge if they are documented. | Knowledge can leave with the supplier unless the contract includes documentation, hand-off, and access to working files. |
| Scaling | Capacity grows through hiring, process improvement, automation, or contractors. | Capacity can often be increased faster, although scope changes normally affect cost and account attention. |
| Exit cost | Reorganisation or hiring changes can be slow and expensive. | Ending a contract may be easier, but transition risk rises if data, documentation, or ownership is unclear. |
The same budget can therefore produce very different operating capacity. A small company with strong product knowledge but no technical SEO capability may gain more from targeted external support than from hiring a generalist. A large company with continuous publishing, proprietary data, complex approvals, and multiple markets may find that weak internal ownership becomes the larger cost even if agency execution is excellent.
Three Operating Models, and Where Each One Breaks
In-house led
An in-house-led model is strongest when search is tightly connected to product, engineering, editorial operations, compliance, or market strategy. The team can build knowledge over time and coordinate with the people who control the website.
Its common failure mode is capability concentration. One experienced person becomes the owner of technical decisions, content history, reporting logic, and vendor relationships. The organisation appears to have an internal capability, but much of it exists in one person’s head. Documentation and cross-functional ownership matter as much as headcount.
Agency led
An agency-led model can work when the company needs expertise quickly, the scope is bounded, or search is important but does not justify building a large permanent team. It is also useful for independent audits and specialist projects that an internal team does not perform often.
Its failure mode is outsourced judgement. If the company cannot explain why a page was created, why a redirect was made, why a metric is being tracked, or what an AI visibility score actually represents, the supplier has become the operating system rather than a specialist resource.
Hybrid
A hybrid model separates ownership from capacity. An internal owner defines priorities, approves claims and changes, controls first-party data, and keeps the decision record. External specialists add technical depth, research, audits, production capacity, localisation support, or temporary expertise.
Hybrid is not automatically the best model. It adds another coordination layer. It works when responsibilities are explicit enough that both sides know who decides, who executes, who verifies, and who documents the result.
How to Evaluate an AI SEO Agency Without Buying the Label
Google updated its guidance on third-party SEO tools, services, and advice in June 2026. It specifically mentions AEO and GEO services and recommends checking third-party recommendations against official Google Search guidance. Google also states that third-party tools do not have access to its internal ranking data and cannot guarantee performance.
Google’s separate guidance on hiring an SEO is still a useful procurement framework. It recommends asking about previous work, industry and market experience, expected results, measurement, communication, and the reasoning behind proposed changes. For AI SEO services, Google now also recommends checking whether advice aligns with its official generative AI Search guidance.
That suggests a more useful evaluation sequence than asking an agency to define GEO:
- Ask what problem they would diagnose first. A credible answer should depend on the site, not begin with a fixed package.
- Ask which claims come from Google documentation and which come from the agency’s own testing or interpretation.
- Ask how they use first-party data. Search Console, analytics, CRM data, logs, and business outcomes should not be replaced by a proprietary visibility score without explanation.
- Ask what they would not change. Good diagnosis includes restraint. Not every site needs a new schema layer, content rewrite, crawler file, or AI-specific workflow.
- Ask how knowledge is transferred. Recommendations, assumptions, implementation notes, measurement definitions, and access should remain understandable to the client.
- Ask how they handle uncertainty. No supplier can guarantee inclusion, citation, or ranking in Google’s generative AI features.
If a proposal depends heavily on special AI markup, llms.txt for Google visibility, content ‘chunking’, or rewriting every page into an AI-specific format, compare those recommendations with Google’s current generative AI Search guidance. Google says those tactics are not requirements for visibility in its generative AI features.
Measurement Is Another Ownership Decision
AI visibility reporting is improving, but that makes measurement governance more important. Google’s Generative AI performance report in Search Console now provides first-party impression data for supported generative AI features in Search, with dimensions including pages, countries, devices, and dates. It does not turn every AI discovery path into a single attribution model.
Third-party tools can still be useful for observing prompts, mentions, competitors, and platforms that Search Console does not cover. The existing MOCOBIN guide to AI visibility tools can help teams assess that layer. But the company should decide which metrics are directional and which are allowed to drive budget or content decisions.
For an agency relationship, measurement ownership should be explicit before work starts. The client should retain access to Search Console, analytics, dashboards, exports, prompt sets, measurement definitions, and the history of major changes. A monthly report is not a substitute for owning the underlying evidence.
Multilingual SEO Changes the Outsourcing Boundary
International SEO adds a second make-or-buy decision because language production and market understanding are not the same capability. A translation team may produce fluent copy without resolving search intent, terminology, page purpose, trust information, or conversion differences in the target market. An SEO specialist may understand technical architecture without being qualified to approve local wording.
For multilingual sites, divide responsibility by the variable that actually changes. Check who owns keyword and intent research in each language, who validates product or service terminology, who decides whether the page structure should differ by market, who reviews local trust and conversion information, and who can approve technical changes such as canonicals, hreflang, templates, or URL architecture.
This is an area where I would avoid a single global template. The operating model should preserve shared technical standards while allowing market-specific evidence to change the content decision. That may require an internal regional owner, a local specialist, an agency, or a combination of them. The job title matters less than whether the responsibility is real.
Run a Capability Audit Before Choosing the Model
Before opening a new role or signing an AI SEO retainer, map the work that already exists. The useful unit is responsibility, not job title.
| Capability | What to verify | Ownership question |
|---|---|---|
| Strategy and prioritisation | Search goals, market scope, page decisions, dependencies, and business outcomes. | Who can decide what should be done first and what should not be done? |
| Technical SEO | Crawling, indexing, rendering, templates, redirects, structured data where relevant, and site architecture. | Who can diagnose the issue, and who can actually implement the change? |
| Content operations | Research, briefs, expert input, drafting, editing, internal links, factual review, updates, and consolidation. | Who is accountable for accuracy and the reason each page exists? |
| AI-assisted workflow | Where generative tools are used, which inputs are permitted, review requirements, and documentation. | Who signs off before generated or transformed material becomes public? |
| Measurement | Search Console, analytics, conversion data, AI visibility observations, reporting definitions, and experiment history. | Who decides whether a change worked? |
| Localisation | Intent, terminology, information needs, market-specific examples, trust information, and technical language routing. | Who can distinguish a translation issue from a market or search-intent issue? |
Once that map exists, the hiring decision becomes easier. Keep responsibilities internal when they depend on proprietary knowledge, approval authority, or continuous cross-team coordination. Buy external capacity when the work is specialised, intermittent, independent, or faster to access than to build. Where both conditions apply, define a hybrid boundary rather than leaving ownership implicit.
The Decision Rule Is Control Over Learning
The durable advantage of an in-house SEO team is not that employees care more, and the durable advantage of an agency is not that outsiders know more. The difference is where learning accumulates and how quickly the organisation can turn that learning into another decision.
If search is central to the business, the company needs enough internal capability to understand the evidence, challenge recommendations, approve risk, and retain the operating history. If specialised work is intermittent or the internal team lacks depth in a specific area, external support can be more efficient than building permanent capacity.
AI search makes that boundary more visible because new terminology and new measurement tools can create false urgency. Google’s own guidance points in the opposite direction: keep the SEO fundamentals, evaluate third-party advice critically, use first-party Search data where available, and do not mistake an AI-labelled tactic for a requirement.
The next decision is therefore not ‘in-house or agency?’ in the abstract. List the responsibilities your search programme requires, mark which ones depend on internal knowledge or authority, and outsource only after the ownership line is clear.
