AI visibility for recruiters: a client reading and a candidate reading

Why an agency needs a client reading and a candidate reading

A plumber is asked about by the person with the leak. A recruiting agency is asked about from two sides that want opposite things from it: the hiring manager wants people, and the candidate wants work. Either may put the question to an AI assistant before contacting anyone. What each side asks, and what an agency's site has to state for each, is on the recruiting and staffing overview; this page is about measuring what the answers do with it.

A single score for both sides hides the gap worth finding. An agency named on its client lines and absent from its candidate lines has a site that talks to employers and gives candidates nothing to repeat, or the other way round. Averaged into one number, that reads as a middling month and points at no page in particular. Kept as two readings, it points at the half of the site that is missing.

So AI visibility for an agency is two readings from one panel: how the answers to clients' questions treat you, and how the answers to candidates' questions do. The verdicts, named and first, and the score, the share of questions that named the agency from 0 to 100 with no weighting, are the ones on our methodology page, applied to each half on its own.

Two ways an answer can serve the wrong side

An answer can name your agency and still fail the person who asked. Two cases deserve a column of their own in the record, because a plain named-or-not verdict scores both of them as a success.

The first is a candidate's question answered with your client pitch. A candidate asks which agencies place [role family] in [city] and what registering involves. If the only text on your site that describes the agency is written for employers (a shortlist in [number] days, a replacement period, a fee model), that is all an answer about the agency has of yours to repeat. The candidate hears why employers hire you, and nothing about registering, pay or whom to call.

The second is a job your agency advertised, credited to the client as the client's own opening. A candidate asks about [role] jobs in [city], and the answer describes the role as [client]'s vacancy, links [client]'s careers page and leaves your agency out. An answer can connect the role to the client only where some page does, such as the client's own listing or a board that names them. The candidate then applies around you, or to a posting that has already closed.

The reverse of the first case is possible too: a hiring manager asks who can supply [number] temporary workers by [day], and the answer quotes your candidate page about weekly pay. How each job page should state the agency's part in a role is covered in what ChatGPT and Perplexity can reach on an agency's site.

Named, recommended and first, read against the side of each line

The three verdicts are the ones every panel records. Each takes a particular shape for an agency, and each is read against the side the line belongs to.

• Named: the agency appears anywhere in the answer, under any form of its name, including as the recruiter behind a job listing the answer mentions.
• Recommended: the answer puts the agency forward for the asker's purpose. On a client line that means as an agency to fill the role; on a candidate line, as an agency to register with or apply through. A mention as a warning, or as one name in a list of employers, is named and not an AI recommendation.
• First: yours is the first agency the reader meets in the answer. Job boards and review sites that appear earlier do not change this verdict; they belong in the cited-page column below.

The side rule settles the hard cases. A candidate line whose answer names you and recites your employer pitch is named, not recommended, because it put you forward to nobody who asked. A client line whose answer names you only as the employer on a job listing is named and nothing more. Written down this way, a verdict means the same thing in both halves, and the halves can be compared.

A panel with client lines and candidate lines

We run twelve questions for a paying client. For an agency, split them between the two sides, and within each side change one thing from line to line: the role family, the industry, the place or the contract type (permanent, contract or temporary). The twelve lines below are a starting shape, with brackets for your own roles, industries and places.

Only lines 6 and 12 carry the agency's name. The other ten are asked by someone who has not heard of you yet, which is where new clients and candidates come from, so read the two named lines apart from them. Once the first reading is taken, the wording stays fixed; setting up prompt monitoring covers the dated panel file, three runs per assistant and how to retire a line without breaking the series.

Weight the halves by where your placements come from. An agency that lives on temporary work may want more temporary lines on both sides; a search firm placing [senior roles] may keep nearly every client line permanent. Write the split down with the panel, because the two readings mean something only against it. The questions each side asks before contacting you, gathered from your own calls, are a different list, set out in what companies and candidates ask about recruiters.

Four columns a recruiting record adds to each answer

The usual record stays as it is: named, recommended, first, who else was named (the start of your share of voice), and the answer kept word for word. An agency adds four columns, filled in by the person who reads the answer, never by a script.

The page-cited column starts empty for a reason. The one large reading we have published counts names, not citations: in our audit's probe, across 9,240 buyer questions, one per business, put once to Claude with web retrieval on, 99.4% of the names given were actual businesses and 0.6% were directories or aggregators. The audit publishes nothing about which pages those answers cited, and it has no separate cut for staffing agencies. How often an answer about an agency leans on a job board is a number only your own record can give you.

Record each board and review site by name, as it appears in the answer. The column is there to trace a fact to the page it came from, not to rank the sites, and a board that cites your posting correctly is a good result.

The employer column has one trap. For temporary work, the agency may be the worker's employer; whether it is depends on the contract and on rules that differ from one jurisdiction to the next. The record does not settle that question. It checks the answer against what your own pages say about who employs the worker, and stating a temp agency's licenses and terms covers where that sentence belongs.

Reading the first result, one half at a time

Read the client half, then the candidate half, then the four columns, and only then the combined score. Each of these patterns points at a different page:

• Named on client lines and missing on candidate lines: check that a candidate page exists at all. Where registering, pay and whom to call are stated nowhere, the candidate lines have nothing of yours to draw on.
• Named on candidate lines only as the employer, or absent while the client is named: read how each job page describes your part in the role, and whether it says so in text.
• Cited pages mostly boards and review sites: the facts about your agency in those answers come from pages you do not write. Set them beside your own pages and publish what is missing from yours.
• A fee or term that matches nothing you publish now: find the page it came from (an old post, a directory entry, a review), correct it there if you can, and state the current term on your own page with its date.

Treat this first reading as a baseline, not a grade. It can show which half of the site to work on first and whose pages the answers draw on. It cannot count the placements or registrations that followed an answer, so ask each new client and candidate how they found you, and keep their replies beside the panel.