An agency founder is a strange person to ask whether you should hire an agency, so let me put my answer first and my reasoning after it: most companies asking this question should start in-house, and a specific minority shouldn’t. The whole method I sell is documented in public, on purpose, so you can check whether I’m describing the work honestly before you believe my answer about who should do it.
The honest version of this decision isn’t about expertise. It’s about which of two things you’re actually short of, and most companies misdiagnose which one it is.
What the work actually consists of
Strip the jargon off and AI search work is two jobs of very different shapes.
The first job is on your own domain: figuring out what your buyers actually ask, publishing pages that answer those questions specifically enough to be quoted, and keeping your site technically readable. This job looks like content work because it mostly is content work. It’s teachable, the feedback loop is visible, and a competent in-house marketer can learn it from public material in weeks. Ours included.
The second job is off your domain: getting the comparison pages, review profiles, community threads, and analyst notes that engines actually quote to say true, current things about you. This job looks like PR because it mostly is PR. It runs on outreach, relationships, and patience, and it’s the part that stalls inside companies, not because nobody knows what to do but because nobody has the standing hours to do it.
Almost every “should we hire an agency” conversation I have is really a conversation about the second job wearing the first job’s clothes.
The case for starting in-house
If you’re a B2B SaaS company under a few million in revenue, here’s what I’d tell you if you weren’t paying me, which you aren’t yet: run the first ninety days yourself.
Start with the audit, which you can do in an afternoon with a free template. Map your buyer questions. Fix what the audit finds on your own site. Publish real answers to the ten questions that matter most. Measure it by hand, three runs per query, once a month. None of this requires a vendor, and all of it builds the thing an agency can’t sell you: your own understanding of where your answers come from.
There are two more reasons the first stretch belongs in-house, and they’re unflattering to my industry. One: the early months of this work produce small, slow numbers no matter who does them, and an agency retainer burning against a flat chart is how companies conclude the whole channel is fake. Two: an agency that starts before you understand the work can’t be evaluated by you. You’ll be grading deliverables by volume, which is exactly the wrong measure, and neither of you will enjoy the quarter.
And if the honest answer to “is this channel even worth it for us” isn’t obviously yes, settle that first. I’ve written the unsentimental version of that ROI case, including the situations where the answer is no.
The line where an agency starts paying for itself
There’s a point where in-house stops being the frugal choice and becomes the expensive one, and it usually announces itself in one of three ways.
The first is when the remaining work is all off-domain and your list of stale sources isn’t shrinking. You know which comparison page is wrong. You’ve known for two months. The email hasn’t been sent, because the person who’d send it has a day job. At that point you don’t have a knowledge gap, you have a capacity gap, and capacity is the thing an agency straightforwardly sells.
The second is when the person doing it in-house is the founder. Founder-led AI search work is high quality and ruinously priced, because the hours come out of sales and product. If you’re personally drafting outreach emails to review-site editors, the agency retainer is cheaper than you are.
The third is when measurement discipline has quietly died. The monthly query runs stopped in March. Nobody noticed until the quarterly review. Consistency is boring, boredom is exactly what you can pay to outsource, and an agency that reports on a fixed cycle keeps the program honest even in the months when nothing moves.
Notice what’s not on this list: secret knowledge. If an agency’s pitch is that they know things about AI engines that can’t be learned in public, they’re describing a moat that doesn’t exist. The real product is senior execution, applied consistently, on the parts of the work your team structurally won’t get to.
The math, in plain numbers
Run the arithmetic before you decide, because it usually decides for you. The in-house version of this program, done properly, is a real fraction of a person: call it thirty to forty hours a month across content, outreach, and measurement once it’s running. If you have a marketer with that capacity and the seniority to write things a lawyer of a buyer might read, in-house is cheap and compounds into institutional knowledge you keep.
If you don’t, price the alternative honestly. Our retainer is $3,000 a month flat, and I publish that number because hidden pricing in this industry does a lot of quiet damage. Compare that against the loaded cost of the hours your team would actually spend, at the seniority the work actually needs, and include the months of ramp-up. Sometimes the spreadsheet says in-house anyway. Fine. The spreadsheet is allowed to win, and you should distrust any vendor who tells you it can’t.
What I got wrong
I used to believe the handoff should happen early, because that’s what was good for me. My pitch logic was simple: the sooner an expert runs it, the less time you waste. Then I took on a client who arrived with zero context, having done none of the work themselves, and I watched the engagement underperform for a reason that had nothing to do with the work.
They couldn’t tell whether I was any good. Every recommendation I made was, from their side, an article of faith. When the early numbers were flat, which early numbers always are, they had no basis for distinguishing “normal lag” from “we hired the wrong person.” The trust ran out before the compounding started, and the engagement ended a month before the results would have arrived. I’ve since watched the mirror image: clients who ran a scrappy version themselves first, hit the capacity wall, and handed it over knowing exactly what they were buying. Those engagements work, because the client can grade the work.
So I changed the answer I give. Do it yourself first, badly if necessary. The education is part of the asset.
How to decide this week
Write down your list of the specific things that would change what the engines say about you. Wrong comparison pages, thin review presence, the segment nobody credible has described you winning. If you can’t produce that list, you’re not ready for an agency, you’re ready for the audit.
If you have the list and the items on it keep not happening, count why. If it’s hours, hire capacity, whether that’s an agency or a focused contractor. If it’s know-how, spend a month with the public material first, ours or anyone’s, and then decide from strength.
And if you get to the point where the capacity you’re short of is exactly the off-domain, source-layer work applied on a disciplined cycle, that’s the work we do and the method we use. You’ll be able to check we’re doing it properly, because you’ll have done it yourself first. That’s not a bug in this plan. It’s the point of it.