The AI visibility report template agencies send clients

Rank reports survived on stability: position seven today is position seven when the client checks tomorrow. AI visibility reporting inherits none of that. The client can rerun any prompt in thirty seconds, will get a different answer than your report shows, and will ask why. SparkToro measured under a 1 percent chance that two identical ChatGPT runs return the same brand list, which means every AI visibility report is one client experiment away from a credibility conversation.

The template below is designed around that conversation. Its two structural rules: every number is a rate over repeated sampled runs, never a single answer, and every claim links to stored raw answers with dates. When the client reruns a prompt and sees something different, the report has already explained why, and the receipts show the distribution their one run came from. Agencies that adopt this posture early convert the volatility from a liability into the reason the client needs them.

Demand is not the problem; G2 found 51 percent of B2B software buyers start research with AI chatbots more often than Google, and clients have noticed. The problem is that most agencies are improvising the deliverable. This is the template to stop improvising with, and if you are still choosing the tooling underneath it, start with AI visibility tools for agencies.

Lead with one number and its movement: share of answers, the percentage of sampled answers to tracked unbranded questions that mention the client. One chart, three or more months of trend, cause markers on the dates fixes shipped. State the sampling underneath in one line: how many questions, how many runs, which engines. That single line of methodology is what distinguishes your report from the screenshot decks the client has seen before, and the construction rules behind an honest headline number are documented in what an AI visibility score actually measures.

Section two reframes the number competitively, because clients experience visibility as a rivalry rather than a rate. Show the same share-of-answers metric for three to five named competitors on the identical question set, as a bar chart this period and a line chart over time. The competitive frame also protects you in down months: when the whole category dips because an engine changed retrieval, the chart shows everyone dipping, and the conversation stays about strategy rather than blame.

The first report of an engagement earns a different opening: the methodology page. One page, plain language, covering the question set and who approved it, the engines tracked, the runs per question, and the branded-prompt exclusion. Clients who read that page once stop questioning the data monthly, and the page doubles as the sales asset that separates your agency from competitors still selling screenshot decks. Every subsequent report can compress it to the single line under the headline chart.

Resist the urge to lead with anything softer. Forrester found 55 percent of buyers compared vendors inside AI tools during their most recent purchase; the client is buying presence in those comparisons, and the first page should say plainly whether they are getting it.

Section three is a table with one row per tracked question: the question text, this period's mention rate, last period's, and a flag for significant movers. Group rows by buyer intent, best-of, alternatives, comparison, how-to, so patterns surface: many clients win how-to questions on the strength of documentation while losing every best-of verdict, and that pattern is the content strategy. Each row links to its dated raw answers. This is the section clients spend the most time in, because it names the exact conversations where buyers do or do not hear about them.

Section four splits the same data by engine. The split earns its pages because engines behave differently: citation analyses find only about 11 percent of domains cited by both ChatGPT and Perplexity, so a client can lead on one engine and be absent from another. Per-engine rows also set up honest scoping conversations about where to invest, since a client whose buyers live in ChatGPT should not fund a Perplexity campaign first. Which engines deserve tracking budget at all is covered in which AI engines matter.

For every lost question, list the domains the engine cited instead, ranked by frequency across the period's runs. This is the section that converts the report from a scoreboard into a work plan, because the cited domains are where the next month's effort goes: a review site the client is absent from, a comparison listicle that omits them, a Reddit thread answering the exact question, a Wikipedia gap.

Structure the section as a ranked table: domain, citation count across the period, the questions it decided, and whether the client has any presence there today. The last column is the one that writes next month's outreach list, and it usually contains more zeros than any client expects. Where a single domain decides several questions, flag it, because one earned placement there outperforms five scattered wins.

The data says this section will surprise clients. McKinsey's AI Discovery Survey found a brand's own site supplies only 5 to 10 percent of the sources AI platforms reference, and University of Toronto research put third-party content at 91 percent of citations. Most clients assume their website is the battlefield; this section shows them the away game, which reframes why the agency's digital PR and off-site work belongs in the retainer.

Section six lists the fixes shipped: pages published with URLs, wrong answers corrected, sources earned, each with its indexing status and, where enough time has passed, the recheck result. Pair each fix to the question it targets so the causality is legible: question lost in June, page published July 3, indexed July 9, mention rate moved from 0 to 40 percent by month end. Not every fix works, and the report should say so; a documented miss with a revised plan builds more trust than a report that only ever celebrates. The craft of pages that actually get cited is its own discipline, laid out in how to write content AI engines actually cite.

Section seven is next period's plan in one page: which lost questions get targeted, which pages are planned, which sources you intend to earn, and what checkpoint the client should expect. Ending on commitments keeps the retainer conversation anchored to the loop rather than to the month's weather.

A note on baselines: the first report of an engagement should present the audit that produced the starting numbers, and low starting numbers are normal. Wellows found over 73 percent of brands have zero AI mentions despite page-one Google rankings. How to run and present that first audit is covered in the AI visibility audit.

Templates fail when they arrive as raw tables, because clients pay for interpretation. Each section deserves two to four sentences of written commentary, and the commentary has a discipline of its own: state what moved, name the most likely cause, and say what you will do about it. "Share of answers rose 6 points, driven by the alternatives page for question 12 getting indexed on the 9th; we are replicating the format for questions 14 and 17" is a sentence a client forwards to their CEO. A paragraph of adjectives about momentum is a sentence they skim.

Calibrate confidence explicitly. AI visibility data supports three strengths of claim: observed (this question flipped, here are the dated answers), likely (the flip followed our page within two weeks and the page is now cited), and possible (category-wide shifts we did not cause). Reports that mark the difference train clients to trust the strong claims because the weak ones were never oversold. This is also where you spend your volatility explanations: a question whose rate moved from 50 to 40 percent on eight runs may have moved by chance, and saying so before the client asks is cheap insurance.

Close the commentary with the one-paragraph executive summary you would want read aloud in the client's leadership meeting, because it will be. Lead with the delta, attribute it, commit to the next move. The discipline compounds: after six months, the stacked summaries read as the story of a program, which is the renewal argument written in advance.

Monthly is the reporting cadence that fits the physics of the channel. Sampling should run weekly or faster, because rates need volume, but reporting faster than monthly amplifies noise into panic: a week of unlucky runs reads as a crisis, and given sub-1-percent run-to-run consistency, unlucky weeks are guaranteed. The exception worth breaking cadence for is a material wrong answer, an engine quoting dead pricing or claiming the client lacks a feature they sell, which justifies a same-week alert and a correction plan rather than a spot in next month's deck.

Deliver the report before the call, then spend the call on sections five through seven rather than reading numbers aloud. The recurring meeting agenda that works: two minutes on the headline and competitive charts, ten on lost questions and the sources behind them, and the rest on agreeing next period's targets, since section seven is a commitment the client should co-sign. Clients who help choose the target questions defend the program internally, because it is partly theirs.

Version the question set deliberately. Questions should change when the client's product or market changes, and each change should be noted in the report the month it happens, because silent question-set edits are the agency-side version of the score inflation this template exists to avoid. A stable set with documented amendments keeps every trend line meaningful back to the baseline month.

The template's data requirements are modest per client but brutal in aggregate: 25 questions, sampled repeatedly across engines, is hundreds of answers per client per month, each needing storage, parsing and competitor tagging. Manual operation caps an agency at one or two clients before the reporting labor eats the margin. The economics only work when the sampling, storage and parsing are automated and the strategist's time goes into sections five through seven, where judgment lives.

This is the loop Reachroller runs. It tracks the unbranded question set on a schedule, stores every raw answer with citations as the receipts sections one through five are built from, and generates the publish-ready fix pages that become section six, each with slug, title tag, meta description and schema markup. Pricing is founder-sized rather than enterprise-shaped: Starter is $29 per month for 400 credits and 25 tracked questions, so an agency can run a paying client's loop for less than an hour of billed time. The honest caveat: it is a young product, ChatGPT tracking is live today, and the remaining engines are rolling out. Plan details are on the pricing page.

Whatever stack you choose, hold the deliverable to the template's two rules: rates over repeated runs, receipts behind every number. The agencies that win this category over the next two years will be the ones whose reports survive the client's own prompt, and the AI visibility retainer is the rare new line item where the measurement discipline itself is the differentiation.