Clients need more than a screenshot showing their brand in an AI answer. They want to know whether visibility is improving, which pages earn citations, where competitors show up instead, and what you plan to do next. This guide is for agency practitioners and marketing leads who put together a white-label AEO report for clients. By the end you will know which sections to include and how to word each one for people who are not specialists.
One idea runs through all of it. White label changes how the report looks, not the standard of proof behind it. Your logo and template go on the front, but the measurement scope, the source labels and the limits of the data stay in. You will need a fixed list of buyer questions, comparable reporting periods, saved AI answers, and access to the client's analytics if you plan to report visits or leads.
The finished report has a cover with the scope, an executive summary, a visibility scorecard, buyer-question findings, cited pages, a competitor view, an outcomes panel, and a list of next actions, with a methodology appendix at the back. You build it in a different order from how the client reads it: the methodology comes first, and the executive summary goes at the front once everything else is done. Not every part needs its own page. A short executive layer with an evidence appendix behind it works better than an unfiltered export.
1. Lock the reporting scope and comparison rules
Before you compare anything, define the questions, engines, markets and counting rules. If you skip this, a change in the report might come from a change in how you measured and not a change in visibility.
Write a reusable scope statement for your white label AEO report. It should hold the client brand and its approved aliases, the client-owned domains that count toward owned-page citations, the reporting and comparison periods with a time zone, and the AI products covered. Keep Google AI Mode and Google AI Overviews apart. Add the country and language settings, the exact tracked prompts grouped by topic and buyer stage, the named competitors, the collection schedule, and your definitions of a valid observation, a failed collection attempt, a mention and an owned-page citation. If a setting or model version is unknown, label it unknown instead of guessing.
Build the prompt set around buyer decisions, not only questions that contain the client's name: non-branded discovery questions, category and use-case questions, comparison questions, and decision-stage questions about fit. Branded questions belong too, mainly to check factual accuracy, but a branded prompt hands the brand to the model, so keep those results separate. Exploratory prompts also live apart from the trend series, and a new topic or engine should show as newly tracked.
For counting, one convention that works is to treat one collected answer to one prompt on one engine at one time as an answer-run. Each period, show the planned runs, the valid observations and the failed ones. Don't count a collection error as the brand being absent, and don't throw out a valid answer just because it had no brand in it. If a vendor uses a different denominator, keep its definition and label any number you recomputed as an agency calculation.
You are done when another analyst could tell which observations are included, redo the main calculations, and explain any change in the sample between periods.
Common mistake: adding easier prompts and then showing the better average as growth. Swapping competitors midway without marking a break, or treating newly tracked prompts as past losses, cause the same problem.
2. Build a visibility scorecard the client can read
Show how often AI answers named the brand, how often they cited its pages, and how those rates moved, with engine-level results visible under any combined headline. If you want the wider picture of how tracking fits together, our roundup of the best AI visibility tools is a good place to look.
Pick a small set of clearly defined metrics. For each rate, show the current value, the previous value, the change in percentage points, and the numerator and denominator. These are reporting conventions, and not every tool calculates them the same way.
| Metric | Recommended definition | Plain-language version | Limit to state |
|---|---|---|---|
| Mention rate | Valid answer-runs naming the brand, divided by valid answer-runs, times 100 | How often AI named you in the answers we checked | Not reach, traffic or market share |
| Owned-page citation rate | Valid answer-runs with at least one citation to an approved client page, divided by valid answer-runs, times 100 | How often AI linked to your own content | Several links in one answer still count as one answer |
| Mention share of voice | Client mention events divided by mention events for the whole comparison set, times 100 | Your share of the tracked brands' appearances | The competitor set decides the denominator |
| Reviewed sentiment | Positive, neutral or negative among the answers actually reviewed | How AI described the brand | Interpretive, and not an accuracy check |
Count a brand once per answer even if its name repeats. Show engine-level results before any overall average. If the baseline is zero, show the absolute change and mark relative growth as not applicable, and if there is no denominator, write "not available," not 0%. Avoid "statistically significant" unless you ran a real analysis, because repeated runs of the same prompt are not independent buyers.
Here is a hypothetical example, with invented numbers only to show the arithmetic. Say you track 40 prompts on two engines with five runs each, which gives 400 valid answer-runs in each period.
| Metric | Previous period | Current period | Change |
|---|---|---|---|
| Answers naming the client | 100 of 400, or 25% | 120 of 400, or 30% | +5 percentage points |
| Answers citing a client-owned page | 40 of 400, or 10% | 48 of 400, or 12% | +2 percentage points |
The mention rate went up by 20% relative to where it started, but it did not go up by 20 percentage points. That is why "visibility increased 20%" can hide a move from 25% to 30%. The clearer sentence is "mention rate rose by 5 percentage points, from 25% to 30%."
You are done when the client can explain each headline number in one sentence and can see the sample size and the engines behind it. The usual mistakes are treating a mention as a recommendation, treating a citation as a click, and showing only a blended score.
DeepSmith's Overview gives you mention rate, citation rate, share of voice, trends and a per-platform view, and the saved answers let you inspect the evidence behind a change. Check the exact metric definition before you quote it next to the conventions above, and remember that which engines you can see depends on your plan.
Client-ready wording: "AI named your brand in 30% of the answers we checked, up from 25%. It linked to one of your pages in 12%, up from 10%. These results describe our tracked question set, not all AI searches."
3. Explain prompt wins, losses and answer quality
Turn the scorecard into buyer questions. Show the ones where the client gained visibility, stayed absent, lost citations or was described wrongly, and pick them because they matter commercially, not because they make the nicest screenshot.
For each highlighted question, include the exact prompt with its topic or buyer stage, the engine and observation dates, and the number of valid observations. Add the mention and citation results, the competitors that appeared in the same answers, a short excerpt with access to the full saved answer, and a line on what it means. Use plain outcome labels: improved, declined, persistently absent, newly tracked, or insufficient evidence. One new citation is something you observed, and it is not automatically a lasting win.
Check factual accuracy and sentiment separately. For accuracy, look at whether the company and products are identified correctly and whether capabilities and pricing are current. A cheerful answer can still hold a damaging error, and a brand that wasn't mentioned shouldn't get a neutral sentiment label by default. Escalate a material inaccuracy on its own instead of waiting for the monthly narrative.
Keep an evidence record for every answer you rely on: the exact prompt, the engine, a timestamp, the full answer, the cited sources and any reviewer notes. When an answer is regenerated, save it as a new observation. Repeating prompts matters here. A 2026 arXiv preprint, Don't Measure Once, found that repeated runs of the same prompt could return quite different source sets. It is one preprint with a limited set of prompts and engines, so use it as a reason to repeat runs and not as a rule about how many to run.
You are done when the client can see which customer decisions the changes concern, and you can produce the underlying answer for any important claim. Watch for cherry-picking one answer, calling a single change a trend, and saying a competitor was recommended when it was only mentioned.
4. Show which pages and sources earned citations
Name the pages AI answers actually cited, explain the questions they supported, and keep client-owned content apart from third-party coverage. Our guide to page-level citation diagnostics goes deeper on what to look at for a single URL.
Build a cited-pages table with the page title, the exact address as captured, the topics and prompts it supported, the engines, the citation count, its share of the client's citation events, the prior-period result and a proposed action. Sort sources into client-owned pages, competitor-owned pages, and third-party sources about the client or category. A review-site citation is evidence of third-party visibility and not an owned-page citation.
Continue the invented example. Forty-eight answers cited a client page, and those answers held 60 answer-page citation events because some cited more than one page.
| Client page | Answer-page citation events | Share of the client's 60 events |
|---|---|---|
| Pricing page | 24 | 40% |
| Comparison guide | 18 | 30% |
| Implementation guide | 12 | 20% |
| Other client pages | 6 | 10% |
The pricing page's 40% is a share of citation events. It is not a 40% citation rate, because the overall answer-level owned-page citation rate is still 48 of 400, or 12%. Always state the denominator, since page percentages measured against client-citing answers can add up to more than 100%.
Then read the pattern. Look for dependence on a single page or engine, decision-stage questions with no suitable client page, and competitor pages that keep supporting questions the client wants to answer. Don't recommend rewriting a strong page only because it is old.
You are done when the client can see which content is being used, for which buyer questions, and what page-level work you recommend. People go wrong by showing only domains, counting URL variants as separate assets, and calling all outside coverage "our citations."
DeepSmith's Pages view shows which client pages AI cites, each page's share of the client's citations and the prompts driving them, and its sources view adds the outside material AI answers reference. Treat these as inputs to your page analysis, not proof that citations caused sales.
5. Benchmark competitors against a named comparison set
Competitor share only means something when the report names the brands, prompts, engines and counting rules behind it. It is not the client's share of the whole AI-search market, and it helps to say that plainly.
State whether the metric is based on mentions or on citations, use the same convention for every brand, and keep the comparison set steady between periods. One answer can name several tracked brands, and each brand gets one event for that answer. The formula is client mention events divided by mention events for all brands in the defined set, times 100. Don't put a mention numerator over a citation denominator.
Using the current period from the earlier example, here are invented numbers.
| Brand | Answer-level mention events | Share of comparison-set events |
|---|---|---|
| Client | 120 | 40% |
| Competitor A | 100 | 33.3% |
| Competitor B | 60 | 20% |
| Competitor C | 20 | 6.7% |
| Total | 300 | 100% |
The client's mention rate is 120 of 400 answers, or 30%, while its mention share of voice is 120 of 300 events, or 40%. Both are correct because they answer different questions. If the set has no events at all, share is undefined and not zero.
DeepSmith supplies a competitor leaderboard and the specific competitor pages cited for your tracked prompts, with performance by engine. Your job is to turn that into a focused explanation of the gap. It does not show another company's private analytics, so don't imply that it does. If you are weighing tools for this kind of client work, the roundup of AEO tools for agencies lays out the options.
6. Separate AI visibility from visits, leads and revenue
Report the business outcomes you can measure, but don't turn citations into clicks or correlation into revenue. Use three labeled layers. The first is tracked AI visibility, meaning the answers you collected for your prompts. The second is native platform exposure, meaning impressions or citations a platform reports for the client's site. The third is website and CRM outcomes, meaning identifiable visits, key events, qualified leads or revenue under a stated attribution method. These layers have different populations and denominators, so don't add them together or suggest they form one fully observed funnel.
Where access allows it, an outcomes panel can show AI-referred sessions by channel, their landing pages, and configured key events such as demo requests. Use "not connected" or "insufficient evidence" where they apply, because a blank integration does not prove zero visits.
Be careful with GA4. Google's default channel group documentation includes an AI Assistant channel for traffic from sources such as ChatGPT, Gemini, DeepSeek, Copilot and Grok, and it excludes Google AI Overviews and AI Mode, which fall under Organic Search. So don't say GA4 has no AI-assistant channel, don't say that channel captures every visit AI influenced, and don't move all Google organic traffic into an AI bucket. OpenAI documents that ChatGPT adds utm_source=chatgpt.com to referral links, which helps you spot inbound ChatGPT traffic, though it doesn't measure every mention or citation. Direct traffic has no clear referral source, so don't relabel it as AI traffic.
Two native reports are worth their own panels. Google Search Console has a Generative AI performance report with impressions from features such as AI Overviews and AI Mode, and values shown as ~ or - become zeros when you download them. Bing Webmaster Tools has an AI Performance report with total citations, cited pages and grounding-query information, but it is sampled and does not show clicks or rankings. A Bing citation count, a Google AI impression and a monitored-answer citation rate are different units, so keep them in separate panels. The setup work is covered in connecting AEO data to site analytics and your CMS, and a comparison of AEO tools that integrate with Search Console can help if you are still choosing. If you want to pull the numbers into your own reporting, the list of AEO tools with an open API shows which ones return live data.
Good aeo white label reporting keeps the scope visible, and you are done when the client can tell apart evidence of being included in an answer, evidence of a visit, and evidence of a business outcome. Calling citation rate a click-through rate and reporting missing analytics as zero are the usual mistakes.
7. Turn findings into owned, testable actions
Every important finding should lead to a decision, an assigned task, or a deliberate choice to wait. Include a dated record of what you completed and a short forward plan, and keep four things separate: the observation, the hypothesis (labeled as uncertain), the action, and the review.
For each action, record the finding, the evidence, the recommended action, the owner, a real due date, any dependency such as product review, the success check, and the date you will remeasure with the same prompt set. Put urgent factual risk, such as a material error about the client's product or pricing, ahead of everything else.
DeepSmith's Opportunity Agents read your AI visibility data or your Content Map and attach the supporting data point to each idea, and chosen ideas can move through New Ideas and Planned Content into Content Studio. That connects report evidence to a content backlog, and it doesn't promise that a published article will earn a citation. When the action involves an existing page, how to check articles for AEO readiness is a natural next read. The roundup of best AEO audit tools covers tools that score pages for citation readiness.
You are done when each priority has an owner, deadline, dependency and review. The usual mistakes are ending on "publish more content," giving the agency credit for every movement, and turning a rise after publication into a causal claim.
8. Brand, check and deliver the report
Apply your branding after the evidence and narrative are sound, so the report is recognizable without hiding what was measured. Use a repeatable template with your identity, the client name, the date range, contact details, consistent terms and a short methodology note. A branded PDF or slide deck is a perfectly good deliverable even when the measurement platform isn't itself a white-label report generator. DeepSmith's visibility and page-level evidence can feed an agency-branded client report, but the documentation doesn't describe native branded PDFs, scheduled client emails or a custom-domain portal, so don't promise those to a client. If you are still choosing a platform, the best AEO tools comparison is the place to start. The AEO tool market map sorts the same field by capability. For multi-brand setups, the best AEO tools for enterprise teams compares workspace isolation across tools.
Keep the executive layer short enough to support decisions and put prompt records and evidence in an appendix. Source labels and limits should survive the rebranding. This summary template is a useful starting point.
"Across [defined questions, engines and market], [metric] changed from [baseline] to [current value]. The most important finding was [supported buyer-question or page-level finding]. [Business outcome or explicit measurement limitation]. Next, we recommend [specific action], owned by [role], for review on [date]."
Before you send, check that every percentage has a denominator, percentage-point changes are labeled, missing observations are not turned into brand absence, the competitor set is named and consistent, and no other client's information or API keys appear in screenshots.
Pro tip: write the executive summary last. It makes it harder to build a positive story before you have looked at the losses, the collection gaps and the attribution limits.
What to do next
For your next reporting cycle, lock the comparison set, replace any unexplained scores with defined metrics, attach saved evidence to the important findings, and give every recommendation an owner and a review date. A white-label AEO report built this way is easier for a client to trust and easier for you to defend. Whichever AEO reporting tool agencies choose, the definitions above are what carry the report.
If you want to see the answers, cited pages and competitor evidence behind your next client report, start a seven-day DeepSmith free trial. The trial doesn't include every engine and doesn't deliver white-labeled reports automatically, so you still assemble the branded document yourself.
Related guides
- How to Use an AEO Platform API for Custom Reporting and BI Dashboards: Pull AEO data through an API into a BI dashboard with a sample schema.
- An AEO Reporting Dashboard for Marketing Leads: Metrics and Layout: Design a dashboard for marketing leads with the right metrics and layout.
- What Are AEO Tools? Categories and What Each One Does: Learn the main categories of AEO tools, including prompt tracking, content optimization, technical and schema checks, and reporting, what each category does, and which problem each one solves.
- What Are Agentic Marketing Platforms for SEO and AEO?: Understand what makes a platform agentic and what it does for SEO and AEO.
- Brandofy Alternatives: 8 Best AI Brand Visibility Trackers for 2026: Eight Brandofy.ai competitors reviewed on published facts and prices: DeepSmith, OtterlyAI, Peec AI, Profound, Scrunch, Rankscale, Goodie, and the Semrush AI Visibility Toolkit, compared on engine coverage, prompt.
