Agency guide

You're a marketing agency with an agency login. This is what you can do, and how to explain the report to your clients.

What you can see

Your login sees only your own tenant: your brands, their reports, and your white-label settings. You never see other agencies, run internals, or any cost data — costs exist only on operator surfaces. Your navigation is Overview · Brands · Reports · Settings · Branding.

White-label settings

These are yours to edit, at Settings → Branding. They apply at render time, so saving a change updates all of your shared reports, including ones you sent last month. Here's what it controls:

Display name — shown in the report header and the "Prepared by …" footer.
Logo — PNG, SVG, or WebP, at most 200 KB, at most 640×160 pixels. Files are validated (real file type, not just the extension) and SVGs are sanitized. It renders at up to 160×40 in the report header.
Accent color — used for links, the headline stat, your client's data series, and the priority column. One kindness worth knowing about: if your color is too light to read, we darken it just for text. Contrast against the report's white and paper backgrounds is checked when you save; a color below the readability floor gets a slightly darkened text version derived automatically. Your logo is untouched, and the settings screen shows you the adjusted swatch. Typography, spacing, and the uncertainty ranges are locked — the honesty of the report is not skinnable.
Hide Mapford branding — removes the "Prepared with Mapford." line from your shared reports. The methodology link stays, and there is deliberately no setting that removes it: a client must always be able to read how the numbers were made. That link is the reason the numbers are worth anything, and it is not a branding choice.

Starting runs, and what a month of allowance means

You can start runs yourself, from any brand page. Two kinds:

Period measurement — the monthly artifact. Its samples are spread across separated days inside the month, because the research this product is built on found that one afternoon's answers are not a month's answers. This is what a client report is.
Snapshot — for a pitch. One day, one pass, at the floor of 8 runs per prompt per engine: the fewest we will report a rate from at all. Everything about it is honest and nothing about it is hidden — it is labelled "Snapshot — a single-day measurement, not a period" on every screen where a rate appears, its ranges print as usual and they are wider, and it never joins a trend line or a "since last time" comparison. Use it to walk into a meeting with a real measurement of a prospect. Do not use it as a month.

Before you start, the preflight shows how many answers the run will sample and roughly how long it will take. It does not show cost: what a run costs us is not a number that belongs on your screen or your client's.

What does constrain you is a monthly allowance your operator sets per agency, on a calendar month. If a run would take the month over it, the run is refused before it starts, with one line: "This run would exceed your agency's monthly allowance; ask your operator to raise it." There are no figures in it, for the same reason there are none in the preflight — ask your operator, who can see them and can raise it.

Once a run finishes, Generate report on the brand page turns it into a numbered, immutable report you can share.

Where to ask. For the two Google surfaces (AI Overviews and AI Mode) the answer depends on the country, so the run form offers United States, United Kingdom, Canada and Australia, and the report records which one it measured. The other engines take no location at all — ask ChatGPT the same question from two countries and it answers the same way — so a run made only of those records no locale, and the report says "not recorded" rather than implying a country nobody measured.

You can still use Request a run with a note if you would rather the operator handled it — for example when a run would exceed your allowance.

From a report page, copy its share link — a long, unguessable URL. No client account is needed. Links can be revoked (and optionally expire); a dead link shows a neutral "this link is no longer active" page with no branding leaked. Shared reports are never indexed by search engines and print cleanly to PDF.

What the report sections mean — for a client

Translate it like this:

The headline verdict ("You appear in 4 of 10 answers…") — how often an AI assistant's answer to a relevant buying question actually names your client, and the shortest path to raising it.
The pages that decide the category — the report leads with the 25 most-cited pages, who controls each (a competitor, an adjacent business, an affiliate, a community, a major publication) and whether your client can reach it. That list is the claim: in most categories a couple of dozen pages carry most of what the assistants read.
Reachability — every page the AI engines read is read by us too, and gets one of four grades: PLACEABLE (a comparison page run by an adjacent business or an independent publisher, a community thread — your client can realistically get named there, typically weeks and modest cost), GETTABLE (big editorial with a submission path, or an affiliate comparison page that lists what pays it — months, relationship-dependent), LOCKED (nothing to buy into: Google's own surfaces, competitors' own pages and their "best of" lists, big directories where everyone is listed anyway), UNKNOWN (we could not read the page; it is reported and never counted as reachable). The bar shows how much of the "citation weight" — the pages engines actually lean on — falls in each grade.
The funnel — three honest steps instead of one blended score: did the assistant look anything up at all? Given that it did, did it name any company in your category? Given that it named the category, did it name your client? Multiplied together, that's the real chance one question produces your client's name. Keeping the steps separate shows where the problem is — no point pitching pages if the engines never activate for these questions.
The targets table — the deliverable. Reachable pages where competitors show up materially more than your client, ranked by how much the engines rely on the page times how far behind your client is. Each row has an estimated cost band and time band. This is the outreach list.
Why every number has a range — "38.0% (32–45, n=210)" means: we asked 210 times; 38% named the client; and the honest error bars say the true rate is somewhere between 32% and 45%. Any tool showing a bare "38%" from a handful of runs is showing noise with confidence. The range is not a hedge — it's the reason the number can be trusted, and it's what makes a before/after comparison meaningful.

Reading the trend, and showing the answers

Two things sit under the report that are worth knowing how to use in a client meeting.

Month by month. Once a brand has two reports in different calendar months, the report and the brand page both carry a series: one point per month for the mention rate and each funnel stage, each with its 95% range drawn as a shaded band, plus placeable share as a plain line. Read the bands, not the dots — where two months' bands overlap, the two months are not distinguishable, and saying "we went from 31% to 34%" across overlapping bands is the thing this product exists to stop you doing. Three rules you will see in action: the axis is calendar months and never anything shorter (a month is the measurement period; a weekly number would be the model's dice, not your work); the line breaks if the prompt set, the engines, the competitor list, the model or the reachability rubric changed between two months, and it names what changed, because those are two different questions and joining them would be a lie; and a month we could not sample enough to report prints no rate at all rather than a dot at zero. Placeable share carries no range on purpose — it is how one month's citations were composed, not something we sampled, so an error bar there would be false precision.

The answers behind the number. This is the one that wins the meeting. From a report, "Read the answers behind these rates" opens every sampled answer, one prompt at a time: which engine and model answered, the date, whether your client was named, the pages it cited with their reachability grade, and the answer text itself. When a client's founder types a prompt into their phone, sees nothing and asks what they are paying for, this is the reply — "across 240 answers your brand appeared in 39; here are five of them" — and you can read the five out loud. Two things to know: single answers vary enormously, which is exactly why we sample hundreds and report a range, so use them as evidence for the rate rather than as findings in their own right; and this view is internal to your login. Shared report links deliberately do not include it, because the raw answers are unedited assistant output that mentions competitors in prose and you should decide what of it your client sees.

Handing it over: PDF and spreadsheets

Every report page — yours and the share link you send a client — has a Download row.

PDF. "PDF" opens a print view of the same report and your browser's print dialog; choose Save as PDF. It is the same report, not a second version of it: same numbers, same ranges, your logo and accent, “Prepared by your agency” in the footer. It always prints on white, whichever theme the reader is using, and it is laid out for paper — a table never splits a row across a page and repeats its header on the next one, a chart never lands half on one page and half on the next. Nothing is stored on our side; the file is yours, on your machine, the moment you save it.

Spreadsheets. Two CSVs from any report, including the client's own share link:

Sources — every page the engines cited, with who controls it, its reachability grade, citation count and weight, how often it named your client and how often it named the strongest competitor, and the estimated cost and time bands for a placement. This is the outreach list in a form you can sort, filter and paste into a plan.
Mention rates — one row per brand with the rate, the sample size and both ends of the range. Four columns, never one: a bare percentage lifted into a slide is exactly what this product argues against, so the range travels inside the file.

A third file, Sampled answers, is on your report only and not on the share link — one row per sampled answer with the engine, the date, whether your client was named, the pages it cited and the answer text. It is the spreadsheet form of "Read the answers behind these rates", and it stays inside your login for the same reason that view does: those are unedited assistant outputs, and you should decide what of them a client sees.

All three open straight in Excel, Numbers and Google Sheets, accents and all.

One thing to never promise: guaranteed AI recommendations. Nobody can guarantee what an assistant says — Mapford measures it honestly and points at the pages most likely to move it. That honesty is the product you're reselling.