How to build and deliver white-label AI search reports to clients in 2026: what to include, the data sources, which tools export today, and a reusable template.
The report a client wants to see for AI search is not the report most tools hand you. Clients do not care that GPTBot fetched 412 pages last month. They care whether AI assistants are starting to send them traffic, whether their site is set up to be read by those assistants at all, and what you actually did about it. A white-label AI search report is the document that answers those three questions with your logo on it, built from data you can defend if the client asks where a number came from.
This is a build guide. I run these reports for client work, and the honest situation as of June 2026 is that no single tool produces a finished, fully white-labeled AI report end to end. You assemble it from a few sources and put it in your own branded container. That sounds like a downside; in practice it works in your favor, because the agency that can stitch logs, a readiness score, and citation data into one clean story is doing something the client cannot buy off the shelf. Below is what to include, where each number comes from, which tools help, and a reusable template you can run every month.
Six sections, in this order, because the order tells a story: here is the demand, here is your readiness, here is the gap, here is the result, here is what I fixed, here is what's next.
AI bot traffic and its trend. Lead with real machine demand. How many times did AI crawlers (GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended) hit the site this month, which pages did they pull, and is that up or down versus last month? This comes from server logs, so it is as accurate as the logs are. It is the one number in the whole report that is not an estimate, which makes it the right thing to open with.
An AI-readiness score. A single 0-100 score for how prepared the site is to be read by AI agents, broken into categories like discoverability, content accessibility, bot access control, and attribution. The score is most useful as a trend line: 54 in April, 71 in June, with the delta attributable to specific work you did. A score that only ever appears once is a vanity metric; a score that moves because of your fixes is proof of work.
llms.txt coverage and the Coverage Gap. Whether the site serves an llms.txt file, how many of its pages are declared in it, and which declared pages no AI bot has actually fetched. That last item, the Coverage Gap, is the interesting one. It is the difference between what you told the assistants exists and what they have bothered to read, and it only exists if you can cross your declared pages against your bot logs. A client understands "you said these 40 pages matter, and bots have ignored 12 of them" without any explanation.
AI referral traffic. Visits arriving from AI assistants, ChatGPT, Perplexity, Copilot, and the like, when you can identify them. Be honest in the report about the limits here: assistants frequently strip referrer and UTM parameters, so a chunk of AI-sourced traffic lands in analytics as "direct." Present this as a directional signal, not a precise count, and note the caveat in the report so the client trusts the rest of your numbers. If ChatGPT clicks keep showing up as direct, that is a known attribution problem with a fix, not a reason to inflate the figure.
Citations and brand mentions. If you run a prompt-sampling monitor, include how often the brand appears in AI answers for its target queries and which competitors get cited alongside or instead of it. This is the share-of-voice layer. Treat the numbers as directional rather than absolute, because LLM answers vary between sessions and most monitors sample a finite prompt set. Our take on why AI share of voice is shakier than vendors admit applies here, so caveat it the same way you caveat referral traffic.
Fixes shipped this month. The section that earns the retainer. A plain list of what you changed: generated and served llms.txt, server-rendered the pricing facts that used to load client-side, opened robots.txt to ClaudeBot, added structured data to the product pages. Tie each fix back to a number that moved where you can. This is what separates a report that justifies your fee from a report that reads like a weather forecast.
Three sources feed the six sections, and the reason agencies struggle is that they live in different tools.
Server logs give you bot traffic and the Coverage Gap. This is the load-bearing source. Every AI crawler hit is a real line in your access logs, identifiable by user agent. A tool that reads those logs tells you machine demand, page-level crawl detail, and, if the same tool also generates your llms.txt, the Coverage Gap. A pure prompt-sampling monitor structurally cannot produce this section, because it never sees your logs. This is the data clients find most convincing precisely because it is not an estimate.
An agent-readiness grader gives you the score. A grader fetches the site's robots.txt, llms.txt, sitemap, and homepage, runs a set of checks, and returns a category-scored 0-100 with itemized findings. The free Crawlytics Agent-Ready Grader scores five categories across roughly 25 checks and hands back a findings list, which doubles as your fix backlog for the month. Run it at the start and end of the reporting period and the delta becomes a slide.
A prompt-sampling monitor gives you citations and share of voice. Tools like Profound, Otterly, and Peec ask LLMs target questions on a schedule and record whether the brand was mentioned and which sources were cited. This is the only one of the three sources that measures presence in answers rather than demand for content or technical readiness. If your client's main concern is "does ChatGPT recommend us," this is the layer that answers it, and it is worth pairing with the other two. The full method is in our guide to tracking AI citations.
No tool I have used as of June 2026 does all three well in one place. The realistic stack is a monitor for citations plus a log-and-readiness tool for the other four sections, exported into one branded document.
Here is where the category actually stands on white-label, with the caveat that these features move month to month, so verify on each vendor's own page before you sell a deliverable around them.
Prompt-sampling monitors are the furthest along on agency packaging. Several offer agency or multi-seat plans, scheduled exports, and in some cases a Looker Studio connector you can style with your own branding, as of June 2026. Profound, Otterly.ai, and Peec AI all market to agencies in some form. What they report is the citations and share-of-voice layer, so they cover one of your six sections deeply and the others not at all. A Looker connector that you rebrand is the closest thing to a true white-label dashboard in this space right now, and it only carries the monitor's data.
Crawlytics covers the other end: bot logs, llms.txt and the Coverage Gap, the readiness score, and AI referral signal. The honest white-label position is mixed. You get a multi-site workspace to manage many client sites, a Custom tier priced for agencies with unlimited sites, CSV-style export, and an analytics dashboard per site. What you do not get yet is a fully white-labeled, embeddable client portal with your agency logo inside the product, that is on the roadmap, not shipped as of June 2026. So the workflow today is to pull the export or screenshot the dashboard and drop it into your branded report. Disclosure: Crawlytics published this post; I have named that gap plainly rather than around it.
The composition tool is your real white-label layer. Because no vendor ships a finished branded AI report, the document itself, your slide template, your Google Doc, your Notion page, your Looker dashboard, is where the white-labeling happens. That is genuinely fine. It means the brand the client sees is always yours, and you are never one vendor's UI redesign away from a report that looks off-brand. Pick a container, build the template once, and the monthly work becomes data assembly rather than design.
Copy this structure into your slide deck or doc and reuse it every month. The fixed order is the point, it makes month-over-month comparison trivial and trains the client to read it the same way each time.
One disclosure line belongs in the footer of every report: which numbers are first-party (logs, readiness checks) and which are estimates or third-party samples (citations, AI referral attribution). Naming that split makes the client trust the whole document more, not less, and it protects you when a monitor's share-of-voice number wobbles between months.
Send the report monthly, on the same calendar day, every time. A 30-day window is the right resolution for this data because AI-bot crawl behavior and citation movement are noisy over shorter spans; weekly reporting mostly reports noise. Quarterly is too slow to show that your work moved anything. Monthly smooths the noise into a trend a client can act on and turns the report into a predictable touchpoint they come to expect.
Most of the template is fixed scaffolding. Three things change month to month, and they are the three you should spend your time on:
The deltas. Every number gets a versus-last-month comparison. The story is never the absolute value, it is the direction. A readiness score of 71 means nothing alone; "71, up from 54 after we shipped llms.txt" means you earned your fee.
The fixes-shipped list. This is fresh every month by definition and is the heart of the document. If a month has a thin fixes list, the report exposes it, which is a feature, it keeps you honest about whether the retainer is producing work.
Next month's plan. Roll the grader's current findings into a short prioritized list. Because the grader returns findings sorted by severity, next month's plan more or less writes itself, and the client sees a continuous loop of measure, fix, re-measure.
Automate the assembly as far as your stack allows. Pull bot-traffic and readiness data from a multi-site dashboard, export the monitor's citation data on a schedule, and keep a single branded template you duplicate each month. The first report takes an afternoon to design; every one after that should take under an hour. When the embeddable branded dashboard ships, the screenshot step goes away, but the template you build now is the same template you will embed later, so the work is not wasted. For more on packaging AI work for agency clients, see our guides on AI visibility tools for agencies and how to prove GEO ROI to clients. Current Crawlytics tiers, including the agency-oriented Custom plan, are on the pricing page, and you can book a demo to see the multi-site view.
Written by Crawlytics Team. Crawlytics tracks AI bots, generates llms.txt, and powers WebMCP commerce, all from one snippet on any stack. See how it works →
Pull data from three sources: a prompt-sampling monitor for citations and share of voice, an agent-readiness grader for the readiness score, and your own server logs for real AI-bot traffic. Export each as CSV or screenshots, then assemble them into a branded document or slide deck in a fixed section order. Repeat the same template every month so trends are comparable. As of June 2026 no single tool produces a fully white-labeled AI report end to end, so most agencies compose the document themselves from exported data.
As of June 2026, several prompt-sampling monitors offer agency-oriented exports or Looker Studio connectors you can rebrand, and Crawlytics offers a multi-site workspace, a Custom (agency) tier, CSV-style export, and an analytics dashboard. None of them ship a fully white-labeled, embeddable client portal with your logo baked in yet. The practical pattern is to export the data and place it in your own branded report. Check each vendor for current white-label features before you commit, since these change month to month.
Include AI bot traffic and its month-over-month trend from server logs, an AI-readiness score, llms.txt coverage including the Coverage Gap (pages you declared that no bot fetched), AI referral traffic, citations or brand mentions if you run a monitor, and a list of the fixes you shipped that month. Open with a one-line summary the client can read in five seconds, and close with next steps. Lead with work done, not just the number.
Yes, if you deliver the report as a branded document or slide deck that you assemble from exported data, which is the standard approach today. A fully white-labeled, embeddable client dashboard that carries your logo inside the tool is on the Crawlytics roadmap and is not available as of June 2026; some monitors offer Looker Studio connectors you can style with your branding. For now, export the underlying data and brand the container yourself.
Monthly is the right cadence for most clients, because AI-bot crawl patterns and citation movement are noisy week to week and a 30-day window smooths the noise into a readable trend. Send it on the same day each month so it becomes a predictable touchpoint. Reserve ad-hoc reports for moments that matter, such as right after you ship a batch of fixes or when a client asks what changed. Weekly tends to over-report noise; quarterly is too slow to show your work.
This page is part of Crawlytics.app. View all pages: llms.txt · llms-full.txt