AI visibility for marketing agencies: named for what, and on whose work

Named, recommended, first: the three verdicts in an agency's terms

Every answer can get three verdicts, and each has an agency twist. Our measurement records two of them, named and first. An agency is named when its name appears anywhere in the answer: in the list, in a closing line of other options, or beside a campaign it made.

It is recommended when the answer puts it forward for what the buyer described, with a reason. For an agency, an AI recommendation comes attached to a brief. Suggesting you for enterprise brand work to a founder who asked for paid social on a small budget recommends you for a job you would turn down. Record the verdict and what it was for.

It is first when it is the first agency the reader meets, which nobody can promise. A campaign credit shows why the verdicts stay separate: an answer about a brand's rebrand can name the agency behind it without recommending it for anything, because nobody asked for an agency. That brand mention still tells you which of your work carries your name.

How a visibility score from 0 to 100 is built is on the methodology page: it is the share of the twelve questions that named the agency, with no weighting. What an agency does differently is what its panel asks and what its record keeps.

A panel built from service, client industry and budget band

Agency questions arrive in three parts, as the page for marketing agencies describes: a service, an industry and some word about money. The panel uses the same three. Each row crosses a service you sell with an industry you have clients in and a budget band you take, typed the way a marketing lead or a founder would type it.

Write the money as buyers do: a figure, or small, startup, enterprise, retainer, project. Take the bands from your own minimums, and put one row below the smallest engagement you accept, as row 8 does, to see whether answers name you to buyers you would turn away.

Weight the panel toward rows 1 to 9, the questions a buyer asks before anyone has referred them to you. Rows 10 to 12 show what answers already attach to your name, which is where the extra fields below earn their keep.

The rest is the same as for any business, and setting up prompt monitoring has it: the file, the runs, the dates, and why the wording is fixed before the first run.

Why the case studies an answer cites describe the agency

An answer that gives a reason for naming an agency has to take it from somewhere. With web search on, the pages an assistant can fetch about an agency are its own and other people's pages about its work, and a case study puts a service, an industry, a client's size and a year in one place, close to the shape of the buyer's question. A homepage that says full-service and names no client gives an answer far less to repeat.

So the case pages describe the agency, whether anyone meant them to or not. An agency that moved from social to performance three years ago, with the social cases still up and the performance work unwritten, has left the old agency in the text. One whose cases all come from a single industry is described by them as a specialist in it, whatever the homepage says.

What a good case page contains is close to what the GEO paper by Aggarwal and colleagues (KDD 2024) tested on the source side: adding statistics, quotations and citations raised a source's visibility in generated answers by up to about 40% on the paper's benchmark. A case study with a measured result, a quote the client approved and a source for each figure has all three. The paper measured visibility on its own set of queries, not on agencies, and not how any assistant arrives at an answer.

Three things weaken a case page: no date, so a campaign from 2019 looks like last quarter's; results only in an image or a PDF, so the text states none; and a client named without current permission. Documenting reference projects covers the write-up, and the article on confidential work covers cases that cannot name the client. Rules on testimonials differ by jurisdiction, so check yours before a client's words go up.

Beyond the verdicts: four fields an agency's record keeps

Each answer is already kept word for word, with its verdicts and every other agency named. An agency adds four fields, filled in from the verbatim text by the person who read it.

The minimum field is the one a site can settle alone. If your page states one minimum and an answer states another, either some other page states that figure or the answer produced it without one. Look for the page first; when AI gets facts wrong covers both cases. If your site states no minimum, there is nothing on your side to hold the answer against.

The client field turns on a question only the agency can answer: may you still name this client? Approval given during an engagement may not have outlived it, the client may have asked to be removed, and a rebrand can leave a case page naming a company that no longer trades under that name. What an agreement allows after it ends is for the agreement and your counsel to say.

Where the answer is no, rewrite the page without the name, with the client's approval of the new text, or take it down, and date the change in the record. If later answers keep citing the work from an awards site or the client's own announcement, ask those hosts, and note that you did.

When an answer names a former employee instead of the agency

Agencies are made of people whose names travel: the strategist with a newsletter, the creative director who judges awards, the founder of a studio that grew out of yours. Any of them can be better known than the agency, and an answer can name the person where a buyer asked for a firm. The fourth field separates three cases.

• Still with you, named without the agency. Check that your team page and the case pages they led name them with the agency, in text, and that their own profiles say where they work.
• Gone, but still placed at the agency. Some page still lists them with you: start with your team page and old bylines, then the event and directory pages from their time with you, and ask those hosts for an edit.
• Gone, with your work credited to their new firm. Check whether your case pages name the agency as plainly as they name the people who did the work.

Erasing anyone is not the fix. A former employee's credit for work they did is true, and a case page can keep it: "led by [name], head of strategy here until [year]". What the agency controls is its own pages: a team page with the people who work there now, cases that name the agency beside the people, and a date on each.

The markup follows the page. A Person node whose worksFor points at the agency belongs only to the people on the current team page; schema.org defines worksFor as the organizations a person works for. When someone leaves, the page and the markup change the same day, and the person stays a separate entity with a name of their own.

What each pattern in the record points to

Read the record by row type before the score. One number across twelve rows can hold steady while the service rows fall and the credit rows rise, and those need different work.

No panel sees every route to a brief. A procurement team's pitch list, a referral from a former client's new employer, a network that passes work between agencies: none of it starts with a typed question. Note how each new inquiry began, with the date, and read that log beside the panel over a year.

Reporting readings like these to your own clients changes the rules, because the agency is then grading work it was paid to do; measuring AI visibility for agency clients covers that. Selling the method is the subject of selling GEO as an agency service.