AI search: advisory, agency or in-house?
Hire in-house if AI search is permanent and full-time. Use an agency if you need execution capacity. Use an advisory if you have a capable team but no senior judgement here. Many companies need none of the three yet.
The three models, honestly
| In-house | Agency | Advisory | |
|---|---|---|---|
| You are buying | Capacity and context | Capacity and process | Judgement and direction |
| Best when | This is permanent and full-time | You lack hands, not direction | You have hands, not senior direction |
| Weakest at | Breadth across platforms early on | Saying you do not need the work | Doing the work themselves |
| Knowledge ends up | With you | With them | With you, by design |
| Ramp | Three to six months | Weeks | Weeks |
| Fails when | You cannot keep them busy | Scope becomes output volume | Your team has no capacity to act |
The case for in-house
One person who knows your product, your category and your buyers will out-perform any outside supplier on the judgement calls, because most of those calls are about your specific situation rather than about AI search.
It is the right answer when the work is genuinely permanent and full-time. That threshold is higher than people assume: monitoring a frozen prompt set and acting on it is perhaps two days a month once established. The rest of a role has to come from adjacent work.
The honest risk is breadth. One person learning eight platforms, five citation-source categories and a measurement discipline from scratch will take two quarters to reach competence, and will be your single point of failure afterwards.
The case for an agency
If the diagnosis is clear and you simply lack hands, an agency is the efficient answer. Content restructuring at volume, technical remediation across a large site, and review-platform hygiene are all execution problems, and execution is what agencies are built for.
The structural weakness is that an agency's revenue is tied to doing work. This is not cynicism, it is incentive design, and it shows up most clearly in the answer to "should we stop?" A model that only earns while producing output is a poor source of advice about whether output is needed.
The practical test is whether they will scope the work down after a baseline. Ask before you sign.
The case for an advisory, and against
An advisory is right when you have a capable team that is executing well on things they understand, and no senior person who has done this specifically. The value is in the judgement: which prompts matter, whether an answer is materially wrong, which of four possible causes is producing your symptom, and when to stop.
This is what we are, so read the following with that in mind.
An advisory fails when your team has no capacity. Strategy handed to a team with no room to act produces a document and a slow-building resentment. If your marketing team is at capacity, buy execution, not direction.
It also fails when the problem is genuinely simple. If your crawler directives are blocking a retrieval bot and your G2 listing is two years stale, you do not need an advisory. You need an afternoon and a checklist, and we will tell you that on the call.
Our position
Roughly a third of the free audits we run end with us saying the problem is not AI visibility, or that the fix is small enough to do yourself. That is a cheaper outcome for us than a badly-scoped engagement, which is why we can afford to say it.
When you need none of them
You have not measured anything yet
Buy nothing until you have a baseline. The four possible causes of low AI visibility have wildly different remedies, and choosing a supplier before the diagnosis is choosing a treatment before the test.
Your organic search is weak
Google's AI surfaces are grounded in Google Search. Fix the foundation first; it is cheaper and better understood.
Your crawler directives are unchecked
An afternoon of work that is occasionally the entire problem. Do this before spending anything.
Your category has little assistant usage
Still true in some industrial and public-sector-adjacent niches. Measure once to confirm, then revisit in six months.
A hybrid that usually works
- 1Get a baseline, from anyone, including yourself, before committing budget.
- 2Fix the cheap, unambiguous things in-house: crawler directives, stale third-party listings, the opening passages of your best-ranking pages.
- 3Bring in senior judgement for the diagnosis and the sequencing, on a defined scope rather than an open retainer.
- 4Use execution capacity, agency or freelance, for the volume work that follows, briefed against that sequencing.
- 5Keep measurement in-house from the start. It is the part that compounds, and outsourcing it means never developing a feel for your own category.
Related questions
How we work, and when we say no
Related answers
How to brief and evaluate a GEO partner
Brief on the outcome and the constraint, not the tactics: what you want to be found for, which competitors you lose to, and what you will not do.
Read the answerDecisionWhat does generative engine optimisation cost?
Across the market in 2026: AI visibility tooling runs roughly £200 to £2,000 a month, freelance GEO £2,000 to £6,000, agency retainers £4,000 to £15,000, senior advisory £8,000 to £35,000.
Read the answerDecisionLooking for a GEO agency or an AI SEO agency?
An agency does the work and is paid for output.
Read the answer