Should You Hire Your Old Search Agency for AI Visibility? An Honest Look (2026)

Your search agency has sent an email about AI. There is a new line item, a new acronym, and a number at the bottom. You want to know whether this is a real service or last year's retainer wearing a hat.

The honest answer is that it is genuinely some of both, and the split is predictable enough to check in an afternoon. A competent search agency is already doing about half of what AI visibility requires, and has no way of telling you whether any of it worked.

So this post draws the line. What their existing work carries over, what it does not reach, the five questions that expose which kind of offer you have been sent, and the cases where keeping them is plainly the right call.

Should you hire your old search agency for AI visibility?

Often yes for the groundwork, and no for the measurement. The page work, the writing, and the work of getting mentioned elsewhere all transfer nearly intact. What does not transfer is any way of knowing whether an assistant names you, because rankings and traffic cannot see that.

That is not a criticism of agencies. It is a description of instruments. A tool built to report where a page sits in a list of links is measuring a different thing from a tool built to report whether a business gets named in a paragraph. Both are legitimate. Only one answers the question you are now asking.

The practical risk is not incompetence, it is substitution. You pay for AI visibility and receive traditional work plus a renamed report, and because the traditional work is genuinely useful, nobody notices for two quarters that the actual question was never measured.

What their work genuinely carries over

More than the market's marketing would suggest. Assistants assemble answers out of pages, so the work of making pages that exist, load, say something specific, and get referenced elsewhere is the same work it always was.

Google's own guidance on creating helpful content reads almost identically to what you would write as a brief for AI visibility: answer real questions, demonstrate first-hand experience, say who you are. An agency that has been doing that well has been building the right thing for a while.

A SEARCH RETAINER, LINE BY LINE, AGAINST WHAT AI VISIBILITY NEEDSILLUSTRATIVE

Technical fixes to your site so pages load and can be read Carries over

Writing pages that answer real buyer questions Carries over

Getting other sites to link to and mention you Carries over

Labeling your facts in machine-readable form Carries over

Knowing which questions buyers ask an assistant, not a search box Partly

Getting you named on the third-party pages assistants lean on Partly

Measuring whether an assistant names you, and how often Does not reach

Correcting an assistant that describes your business wrongly Does not reach

Read it as a scope document. The top four are the same job under a different heading. The middle two need knowledge of how buyers phrase things to an assistant. The bottom two are not in a search retainer at all, and are the ones worth asking about by name. The split is illustrative, from how these engagements are usually scoped.

AnswerHalo answerhalo.com

CLICK TO ENLARGE

Written out, because nothing here may live only inside a figure. Technical fixes, writing pages that answer real questions, earning mentions on other sites, and labeling your facts in machine-readable form all carry over. Knowing which questions buyers put to an assistant, and getting you named on the third-party pages those assistants lean on, carry over only partly. Measuring whether an assistant names you, and correcting an assistant that describes you wrongly, do not reach at all. That split is illustrative.

What it does not reach

The measurement, and everything downstream of it. Nobody can tell you which of your buyer questions you are losing unless somebody asks those questions and reads the answers, and that is not a byproduct of any search tool.

Two things follow from that. You cannot prioritize, because you do not know which questions you lose. And you cannot tell whether the work is working, because the number going up in your report is a proxy for a different outcome, which is the gap covered in ranking on Google but missing from AI answers.

The second thing out of reach is correction. When an assistant describes your business as something you stopped doing, there is no form to submit and no account manager to call. Perplexity's help center describes answers being composed with sources attached each time they are asked, which is the reason: there is no stored profile of you to edit, only the web it reads at the moment of the question.

Neither of those is a failing an agency can fix by trying harder. They are outside the instrument. What you want to know is whether the agency knows that, and says so.

Five questions to ask before you sign

All five are answerable in a sentence by an agency doing real work, and all five are awkward for one that has relabeled a retainer. Ask them in writing.

The last row is the one that decides it. Anybody willing to guarantee you a spot in an assistant's answer is describing a mechanism that does not exist, and that single claim tells you more about the offer than the other four answers combined.

The fourth row matters nearly as much. Without a dated baseline there is no honest way to attribute anything later, and establishing one is cheap. It is the same discipline as the comparison between monitoring tools and done-for-you services: the deliverable that matters is the one that tells you the score changed.

When keeping them is the right call

If the groundwork is genuinely undone, keep them and let them do it. A site that loads slowly, says little, and is mentioned nowhere else does not have an AI visibility problem yet. It has the older problem, and the older work is the correct first spend whatever anyone calls it.

Keep them too if the relationship is good and they are honest about the boundary. An agency that says plainly "we will do the page and mention work, we do not measure assistant answers, here is who does" is being more useful than one promising the whole thing. That is a partner, and they are not common.

Leave, or split the work, when the report does not change. If three months in you are still reading rankings and sessions, you are paying for traditional work at a new price, and the question you started with has never once been asked out loud.

One more honest note, since this post is about spending decisions. Some of what gets sold as AI visibility is just the old craft done well, and the research supports that: Princeton's work on generative engine optimization found that citing sources and quoting authority raised how often a page was used in generated answers. Good writing with real references has always been the job, and it does not need a new acronym to be worth paying for. The judgment call about whether you need outside help at all is in what generative engine optimization is, and whether you need it. And if a monthly arrangement is the wrong shape for you entirely, who sells a one-off audit instead lists the alternatives with the prices they publish.

Our free check runs 25 real buyer questions from your niche at $0 and emails a plain-English readout within the hour, which gives you a dated baseline to hold any agency to. It runs as soon as you ask for it.