How to Package AI Visibility Reporting for Agency Clients

Citation frequency is the count of times a client's domain appears in AI-generated answers for a defined query set over a reporting period. Citation position records whether the client appears as the first, second, or third attributed source - this matters because AI engines typically surface two to four sources, and the first-cited source tends to get the most follow-through attention. These two numbers form the foundation everything else in the report builds on.

AI share of voice (aSOV) is the proportion of AI-generated answers in a topic cluster that reference the client's brand, expressed as a percentage. If you track 20 queries in the "project management software" cluster and the client appears in 8 of the 20 AI answers, their aSOV is 40%. This is directly analogous to traditional share of voice in paid search, which makes it immediately legible to a CMO who already tracks SOV elsewhere. Calculate it monthly so the client sees a trend direction, not just a point-in-time snapshot.

You do not need a paid tool to build this: pull the cited URLs from your prompt-audit log, run each through Google's Rich Results Test to check for schema, check the last-modified signal in the page source or via a curl -I request, and score each page on each of the three dimensions. A page scoring low on any single dimension is a fix priority for the next sprint.

Content freshness delta measures the gap, in days, between your client's most recently updated cited page and the most recently updated competitor page appearing in the same AI answer. A client page last touched six months ago sitting next to a competitor page updated two weeks ago is one of the most actionable numbers a report can surface, because it points directly at a fix with an obvious owner.

The scorecard format that tends to earn the highest client retention is a one-page summary that shows trend direction, not just current state. The reason is structural: a ranking report answers "where are we," while a citation gap report answers "why is our competitor there and we are not" - and the second question is the one that drives budget decisions. Lead the scorecard with the trend line, then the gap, then the fix.

Build it in a spreadsheet shared directly with the client. One column lists the ten to twenty tracked queries. Three more columns capture citation presence in ChatGPT, Perplexity, and Google AI Overviews respectively - cited, absent, or cited-but-not-in-the-top-two. Another column names the top competitor cited in each answer, and a final column records the likely reason (schema present on the competitor's page, fresher content, a more specific entity match). That final column is the difference between a report the client reads once and one they actually bring to their content team.

Run the prompt audit manually on a fixed day each month - the first Monday works well, since it gives you the weekend's indexing cycle. Use a clean, logged-out browser session to avoid personalization artifacts. For ChatGPT, use browsing mode and note the source URLs it surfaces. For Perplexity, use default search mode and export the source list. For Google AI Overviews, run the query from a session geolocated to the client's primary market. Log every cited URL before you start interpreting anything - raw data first, analysis second.

Keep the scorecard's summary section to three numbers: current aSOV, the change from last month in percentage points, and the count of queries where the client moved from absent to cited. Those are the three numbers a CMO is likely to repeat in their next board meeting. If you add narrative, keep it to three sentences: what changed, why, and what happens next. Anything longer tends not to get read.

A competitive citation gap is a specific, answerable question: this query returns a named competitor in the AI answer - why them and not us? The answer is almost always one of four things: the competitor has schema markup the client lacks, their page was updated more recently, their content answers the query directly within the first hundred words, or their domain has more topical depth on the specific subtopic. Identifying which of the four applies matters because each maps to a different fix owned by a different person on the client's team.

Google removed FAQ rich results from Search in May 2026 and deprecated HowTo rich results in 2023. Keep visible FAQs and steps when they help readers, but do not present either markup as an AI-citation or rich-result lever.

To diagnose a freshness gap, compare Last-Modified signals: run curl -I against the competitor's cited page and the client's equivalent page, and compare the dates. If the competitor's page was touched recently and the client's has sat untouched for months, frame it plainly in the report - "competitor updated this page recently; our equivalent page has not been touched in months; the AI engine is citing the fresher source" - which is the framing that tends to get a refresh actually prioritized.

To diagnose a topical-authority gap, count how many pages the client has published on the specific subtopic versus the competitor. A single client page against a dozen competitor pages on the same subtopic is a real depth gap that an AI engine can pick up on. The fix here is a content-cluster build - a longer-term editorial task, not a quick patch.

Position AI visibility reporting as its own line item, not an add-on buried inside a general SEO retainer. The reason is contractual clarity: when a client can see "AI Visibility Monitoring" as its own line on an invoice, they can budget for it, defend it to their own finance team, and choose to cut or expand it independently of everything else. Bundled into a general retainer, it becomes invisible - and what is invisible is the first thing cut when budgets tighten.

Anchor pricing to query-set size and platform coverage, not to hours. A reasonable structure: a starter tier covering roughly ten tracked queries across ChatGPT, Perplexity, and Google AI Overviews, delivered as a monthly scorecard with a short review call; a growth tier covering roughly twenty-five queries that adds competitive gap analysis for the top few competitors and a quarterly schema audit; an enterprise tier covering fifty or more queries that adds longer-horizon aSOV trend reporting and refresh recommendations mapped to specific gap causes. A fixed deliverable list per tier is what makes the service sellable without constant scope negotiation.

The fastest path to expanding a retainer is a pilot. Pick one client with an active content team, run the manual scorecard for ninety days at no charge or a reduced rate, and document the citation-frequency change. A specific result - moving from zero to several citations across a named set of queries in ninety days - is more persuasive in a sales conversation than any general pitch about AI search trends, because it is a concrete, checkable outcome rather than an industry claim.

When a client pushes back with "we only care about traffic and leads, not mentions," the direct response is that AI citations are an upstream driver of exactly that traffic. Pull the client's own referral-source breakdown in Google Search Console, filter for AI-adjacent referrers, and ask them to account for the traffic they cannot otherwise attribute. That unexplained traffic is the business case, made from their own data rather than an industry statistic.

The most common objection is that LLM answers are too volatile to track. That objection is partly right and partly wrong. It is true that an individual AI answer can vary by session, account history, and real-time retrieval. It does not follow that trend tracking is useless. Running the same query set from a clean session on the same day each month measures the baseline retrieval behavior of each platform, not every possible answer variant, and that baseline is stable enough to show a real trend direction across monthly intervals - which is what a client report actually needs.

Control for session variance with a consistent protocol: a logged-out browser, the same geographic IP each month, and platform-specific settings held constant (disable ChatGPT's memory and start a fresh conversation each time; use Perplexity's default search mode without a logged-in account; match the client's primary market for any geolocated query). Document the protocol in the report's methodology footnote so the client understands why the numbers are comparable month over month.

Cross-validate citation claims against Google Search Console. If a prompt audit shows the client cited in a Google AI Overview for a specific query, there should be a corresponding impression movement in Search Console for that URL around the same date. If the impression data and the citation log consistently disagree, the audit protocol has a gap somewhere - most often a regional or personalization artifact - worth tracking down before you report the number to the client.

Recheck the tracked query set every quarter. A query that triggered an AI Overview answer in one quarter may switch to a standard organic result in the next if the underlying system reclassifies its intent, and new queries in the client's topic cluster may start triggering AI answers where they did not before. A quarterly refresh - adding a handful of new queries, retiring ones that stopped triggering AI answers - keeps the tracked set representative. Log the reason for every addition and removal so the client can see the methodology is maintained, not manipulated.