← Blog

How Agencies Run a Weekly AI Visibility Loop

One mission per week: Evidence → Action → Recheck → Result — how to turn a finding into a next step, paste a client update, and report only measured wins.

Agency operators8 min readNed, Prescy

Boutique agencies do not need another dashboard tour. They need a repeatable week: one gap, one brief, one recheck, one honest client note.

This guide mirrors what Prescy’s Home, Findings, and copy-out cards already do — so your team can run the loop without inventing a process around the product.

Key takeaways

  • One mission per week: Evidence → Action → Recheck → Result.
  • Use Your next step on a Findings row and Paste into a client update on Home — do not rewrite the measurement in score language.
  • Two weekly runs are the earliest useful before-and-after. Do not sell a one-run wobble as a win.
  • If Monday shows no clear change, keep the baseline and change the lever only after a second quiet Monday.

The weekly loop

  1. Evidence — open Home. Read the named-in-X-of-Y headline, the rival, and the cited source on the top Confirmed finding.
  2. Action — brief the writer (or the client’s owner) with the exact URL and what “done” means: a live public page, then mark the source watched.
  3. Recheck — wait for Monday. The baseline stays locked while you wait.
  4. Result — read improved / declined / mixed / no clear change, then take the suggested next tactic.

That is the whole operating system. Charts and history stay available for deep dives; they are not the weekly meeting.

How to turn a finding into a next step

On Findings, each row shows Your next step — a plain move based on the page ChatGPT actually cited for that losing question, not a generic "rewrite something". It reads differently depending on who owns that page:

  • Your own page — rework it so the recommendation-shaped answer (who it is for, what it replaces, one proof point) is near the top. Fix the page, do not add a new one.
  • A competitor's own page — you cannot edit it. The step is your own comparable page plus one neutral page that ranks for the question — a roundup, a review site, a directory.
  • A review site, roundup or directory — a page you can get onto. The step is the first listing or pitch action, with a rough time expectation.
  • No cited page — ChatGPT answered from memory. The step is making sure your own page for that question exists and can be crawled.

Copy as a task copies the same text with the URL and an owner line, for a writer or teammate. Done still means the same thing: a live public URL, then your name appearing in a later weekly recheck — not the pitch email, and not the first quiet recheck.

How to paste a client update

On Home, open Paste into a client update. Copy the note into Slack or email. A non-Prescy reader should understand it in sixty seconds:

  • Named in X of Y (never a proprietary score).
  • The buyer question and a short verbatim excerpt.
  • The cited domain and whether it names the brand.
  • The next action and when it will be rechecked.
  • Your optional one-line operator note.

Lead with Confirmed work. Keep Emerging findings out of the headline. If nothing moved on the first Monday, say so — quiet weeks stay quiet.

Anonymised case: question → URL → edit → recheck

A B2B SaaS client was never named on “best [category] tools for mid-market teams.” ChatGPT cited a roundup on a trade publication. The agency:

  1. Copied the next step — the cited roundup URL and the missing comparison criteria (real customer size, pricing band, security note).
  2. Pitched an updated listing (real customer size, pricing band, security note) — no new microsite.
  3. Marked the source watched when the live URL shipped.
  4. Locked the baseline (0 of 7) and waited for Monday.

First Monday: still 0 of 7 — no clear change. They left the baseline locked, worked a parallel on-site FAQ finding, and waited a second Monday. Second Monday: named in 2 of 7. That is the earliest honest “it moved” story — not the pitch email, not the first quiet recheck.

What not to do

  • Report five engines as if Monitor scores them all today — live measurement is ChatGPT and Perplexity.
  • Translate X of Y into a vanity percentage for the slide deck.
  • Brief three writers on three Emerging rows in the same week.
  • Call a one-run bump a win when nobody shipped a watched URL.

Track it automatically

Want to track this automatically?

Prescy shows who AI recommends, why competitors may win and what to change — starting with a free private AI Search Report.

See who ChatGPT recommends →