TL;DR
Google's June 3, 2026 Search Console Generative AI report gives impressions data for AI Overviews but excludes click-through rates, query-level prompts, and any data before May 18 — meaning you cannot report a CTR or compare pre-May numbers. GA4's new "AI Assistant" channel (launched May 13) auto-tags ChatGPT, Gemini, and Claude but misses Perplexity, which lands in generic Referral unless you build a custom regex channel, and the change isn't retroactive, so month-over-month comparisons will show artificial shifts. The article's template forces a fourth section — "Assumptions & Unknowns" — to explicitly disclose these gaps instead of burying them in footnotes.
The bottom line: copy the four-section structure into a doc, fill it monthly, and keep the unknowns section honest — hiding the channel's immaturity behind confident numbers will eventually get someone in trouble.
Most "AI search visibility" content on the web is philosophy: why answer engines matter, why citations are the new backlinks, why you need an AEO strategy. Useful, but none of it produces the thing a marketing lead actually has to hand their VP on the first Monday of every month — a report with defined sections, real rows, and a clear line between what happened and what you're guessing.
This is that artifact. It's built around four sections: observed AI answers and citations, referral traffic data, content changes shipped, and — the section most reports skip — assumptions and unknowns. Copy the structure below into a doc, fill it in monthly, and keep the unknowns section honest. That last part matters more than it sounds like it should, because the tooling landscape for this channel is genuinely immature right now, and a report that hides that immaturity behind confident-looking numbers will eventually get someone in trouble.
Why the unknowns section isn't optional
Direct answer: Two platform changes in mid-2026 make this template possible at all, and both come with real gaps you need to disclose rather than paper over.
Google shipped a dedicated Generative AI performance report inside Search Console on June 3, 2026, giving site owners impressions data for how their pages surface inside AI Overviews and AI Mode, filterable by URL, country, and device. It's a real improvement — but per Google's own announcement, the report excludes click data, doesn't expose the actual prompts that triggered a citation, and only tracks data from May 18, 2026 forward with no historical backfill. Any report section built on this data has to say, explicitly, "no CTR, no query-level detail, no pre-May data" — not bury that caveat in a footnote.
Google Analytics added a similar half-step. As of May 13, 2026, GA4's Default Channel Group includes a native "AI Assistant" channel that auto-tags qualifying sessions with medium ai-assistant, per Google Analytics' own changelog. It recognizes referrers from assistants like ChatGPT, Gemini, and Claude out of the box. It does not cover every AI referrer — Perplexity traffic, notably, still lands in generic Referral unless you build a custom channel group with a regex rule to catch it, and the native channel isn't retroactive, so month-over-month comparisons that straddle the May 2026 rollout will show artificial channel shifts that aren't real traffic changes.
Both gaps are exactly why this template forces a fourth section instead of letting the first three sections imply more certainty than the data supports.
The template
| Section | What to log | Where the data comes from | Example row |
|---|---|---|---|
| 1. Observed AI answers & citations | Query/topic monitored; AI surface (Google AI Overview, AI Mode, ChatGPT, Gemini, Claude, Perplexity, etc.); cited (Y/N); URL cited; any position/rank signal available | Search Console's Generative AI performance report (Google surfaces only) + manual or tooled checks against chat-based assistants | "expense management software for 50-person teams" — Google AI Overview — cited — /product/expense-management — no position data (GSC doesn't expose it) |
| 2. Referral traffic data | Sessions via GA4's native AI Assistant channel; sessions via custom AI regex channel (catches Perplexity, others); AI-attributable share of Direct (flagged, not claimed); engagement rate/avg. engagement time vs. site average | GA4 Default Channel Group + a custom channel group built per Google's own guidance | AI Assistant channel: 142 sessions, 61% engaged sessions, avg. engagement 2m14s (site avg: 1m38s) |
| 3. Content changes shipped | Page/URL touched; change type (new page, schema/entity markup, FAQ block, direct-answer rewrite, freshness update); date shipped; the specific hypothesis behind the change | Editorial calendar / CMS changelog | /blog/switching-payroll-providers — added FAQ schema + a direct-answer opening paragraph — Aug 6 — hypothesis: improve odds of AI Overview citation for "how to switch payroll providers mid-year" |
| 4. Assumptions & unknowns | The assumption being made; why it can't be verified this cycle; what evidence would confirm or kill it | Analyst judgment, stated plainly | "The 18% rise in unattributed Direct traffic this month is AI-referral leakage" — can't verify because GSC excludes clicks and GA4's AI channel isn't retroactive across the comparison window — will re-test once we have two full clean months post-rollout |
Filling in Section 1: observed answers and citations
Log every query or topic you're actively tracking, not just the ones where you got a hit — a miss is data too, and it's the difference between "we're not visible here yet" and "we don't know." For Google's own AI surfaces, Search Console's Generative AI report is now the source of truth for impressions, but remember it only started collecting data from May 18, 2026 and carries no click-through numbers, so don't report a CTR for this channel — you don't have one. For chat-based assistants (ChatGPT, Gemini, Claude, Perplexity), you're still checking manually or through a monitoring tool, since none of them publish a citations API; note the check date next to each row, because these answers are non-deterministic and can change citation between two prompts run minutes apart.
Filling in Section 2: referral traffic
Report GA4's native AI Assistant channel number first — it's zero-config and, per Google, automatically classifies sessions the moment the update reached your property. Then report your custom-channel-group number separately, and label it as such, since it's catching a different (and currently non-overlapping) set of referrers, chiefly Perplexity and any assistant Google hasn't added to its default list yet. Do not merge these two numbers into one "AI traffic" figure without a footnote — a reader comparing month to month needs to know if the increase is real traffic growth or just more of the same sessions getting correctly bucketed as Google's classifier improves.
Where this section tends to go wrong is Direct traffic. Industry analysis — including Conductor's 2026 AEO/GEO benchmarks report, based on 3.3 billion sessions across more than 13,000 enterprise domains — found that a large share of brands cannot properly attribute AI referral traffic at all, in part because assistant sessions frequently arrive with no referrer header and land in Direct by default. That's a real, well-documented gap. But it is not license to relabel your Direct traffic as AI traffic in a report. State the gap in Section 2 if you want ("Direct traffic is up X% this month; some portion is likely AI-referral leakage per industry research, but we cannot isolate it"), and put the actual unverifiable claim in Section 4 where it belongs.
Filling in Section 3: content changes shipped
This is the section most reports get right by default, because it's just a changelog. The discipline to add: tie every change to a specific hypothesis about AI visibility, not a generic "improved SEO" note. "Added FAQ schema to improve AI Overview citation odds for X" is falsifiable next month against Section 1. "Improved SEO" is not. If a change was made for traditional organic ranking reasons and AI visibility is a secondary hope rather than the actual goal, say so — it keeps you from later crediting an AI citation to a change that was never really targeting AI answers in the first place.
Filling in Section 4: assumptions and unknowns
Direct answer: This is the section that makes the whole report trustworthy, and it's the one most templates skip because it doesn't look like progress. It should. Structure each row as assumption → why it's unverifiable this cycle → what would confirm or deny it. Common candidates for any team running this report through the rest of 2026:
- Whether a citation in an AI Overview or a chat answer actually drove the session you're crediting to it — GSC's report gives impressions, not clicks, and chat assistants generally don't pass query-level referrer data at all.
- Whether a Direct traffic increase is AI-referral leakage or something else entirely (a PR mention, a returning-visitor spike, an app deep link).
- Whether the AI Assistant channel's month-over-month trend is real growth or a rollout artifact, since Google's classifier only started applying retroactively-blind as of May 13, 2026 — any comparison that spans that date is comparing two different measurement regimes, not two months of the same channel.
- Whether losing a citation in one assistant reflects an actual content or authority problem, or normal churn — answer engines resample sources between runs, and a single miss is not yet a trend.
None of this is a reason to withhold the report. It's the reason to ship it with a section that says, in writing, what you don't yet know — so the next month's report can either confirm or retire each assumption instead of quietly restating it as fact.
Building this into a monthly habit
The structure above deliberately mirrors what mainstream marketing-reporting practice already recommends for any channel: define scope and comparison basis up front, break performance out by channel with consistent fields, and close with what's next rather than just what happened — a pattern documented across current monthly marketing reporting guidance and marketing report format examples. AI-search visibility isn't exempt from that discipline just because the underlying platforms are new — if anything, the newness of Search Console's Generative AI report and GA4's AI Assistant channel is exactly why the unknowns section needs to be a permanent fixture, not a temporary hedge you drop once the tooling "catches up." Given the pace of change this year, it probably won't catch up on your reporting schedule — so build the honesty in from month one.
Sources:
- Introducing Search Generative AI performance reports in Search Console — Google Search Central Blog
- What's new in Google Analytics — AI Assistant channel
- Custom channel groups — Google Analytics Help
- The 2026 AEO/GEO Benchmarks Report — Conductor
- Monthly Marketing Reports: Definition, Examples, and Best Practices — Sona
- Marketing Report Examples: 8 Client-Ready Reports for 2026 — AgencyAnalytics
