Should we hire a GEO agency or buy software and handle AI search in-house?

We are an agency, so you might expect us to say hire an agency. We do not say that to every prospect, and the reason is worth explaining. Some teams are better served by software and one capable person. Others need people. Many need both.

Here is how we think about it, based on how we run programs.

Separate the work from the measurement

The decision gets simpler once you see that there are two jobs. One is doing the work: choosing prompts, writing and refreshing content, fixing crawler access, earning mentions on third-party sites. The other is measuring whether any of it moves AI visibility.

Software is good at measurement. People are good at the work. Trouble starts when a single party owns both, because nobody can check the result independently.

So our advice is nearly always the same: whoever does the work, the client should own the measurement platform and have their own login.

What software cannot do

A platform runs prompts, records answers and reports visibility, citations, crawler hits and traffic. It does not decide what your brand stands for, build relationships with the publishers that AI engines cite, or write a point of view worth citing. It also cannot guarantee that an engine will name you. Answers change between runs, models and users, and any provider who says otherwise is selling something.

What an agency cannot do

We cannot give you a ranking guarantee either. What we can provide is experience, capacity and judgment: which prompts matter, what a good answer page looks like, which technical issues block AI crawlers, and how to prioritize when everything looks urgent. We also bring pattern recognition from running many programs, which a first-time in-house owner does not have yet.

When we tell people to stay in-house

You already have a marketer who understands your product and audience, your budget is limited, and you are still testing whether AI search matters for your category. In that case, buy a platform, start with a small prompt set and learn the baseline. Promptwatch has a free Explore plan limited to ChatGPT and 10 prompts, which is enough to see where you stand. Paid brand plans start at Essential, $95 a month.

When we tell people to hire help

You have a large content backlog and no capacity to clear it. You operate in many markets or languages. Your leadership wants a program running within weeks rather than quarters. Or the work needs specialists, such as technical SEO, digital PR and editorial, that you do not have.

What we run our programs on

We run client programs on Promptwatch. The reason is practical: it covers the whole chain in one place. Prompt tracking with volumes and difficulty. Citation analytics that show which pages and domains the engines pull from. IP-verified AI crawler logs through Agent Analytics. Visitor analytics that attribute AI-referred traffic and conversions. And content tooling, including content gap analysis and Content Refresh.

Most alternatives are trackers: they tell you whether you were mentioned. They do not show why, where the engine found its sources, or what traffic it produced. That diagnostic layer is what makes the work efficient. Without it, an agency is guessing and so is an in-house team.

A note on plans: crawler logs start on Professional at $245 a month, not Essential. Agency-style plans start at Kick-off at $199 a month and include unlimited projects, 10 seats and crawler logs, though they do not include AEO articles or the automated Content Agent.

A split that works

The client buys the platform and owns the login.
We set up prompt panels by topic, type, country and language, with the client's team.
The in-house lead owns strategy, product knowledge and approvals.
We handle production, refreshes, technical fixes and outreach.
Everyone reviews the same dashboard monthly, and the Action Board keeps priorities visible.
After two quarters, revisit who does what. Some clients bring more in-house. Some add scope.

Questions to ask any agency, including us

Which platform do you measure in, and will I get my own login?
What happens to the data and the setup if we part ways?
Which deliverables are in scope, and how often do we meet?
How will you report when the data is noisy or a result is unclear?
What will you not promise?

A good answer to the last one is a short list. If the answer is "nothing," keep looking.

The honest summary

Own the measurement, rent the capacity you lack. If you are starting from zero, begin with Promptwatch and a small prompt set, then decide how much help you need once you can see the gaps.