Building a channel mix scorecard: how to know which of your marketing channels are actually earning AI citations

Key takeaways

• A channel mix scorecard for AI citations needs three layers: citation data (who's actually being cited), crawler data (who's reading your content), and referral data (who's sending traffic) — most teams only track the third one, and it's the least reliable.
• Reddit's share of ChatGPT citations fell from roughly 4% to 0.5% almost overnight on August 14, 2026, which is a reminder that any scorecard built around a single channel is fragile.
• Product pages now dominate ChatGPT citations at 32.8% (July 2026), nearly double their March share — if your content mix skews toward blog posts and listicles only, you're optimizing for a shrinking slice of the pie.
• Social channels behave completely differently per engine: ChatGPT is basically a Reddit machine, while AI Overviews and Grok lean on YouTube. There's no universal "social wins" row on this scorecard.
• Standard web analytics (GA4, Adobe) systematically undercount AI referral traffic because of UTM inconsistency and stripped referrer headers, so a scorecard built only on Google Analytics will lie to you.

Why your existing channel report doesn't tell you this

Most marketing teams already have a channel mix report. Direct, organic, paid search, paid social, email, referral — the usual suspects, sliced by traffic and conversions. It's a fine report. It's also almost useless for answering a question that's become unavoidable in 2026: which of these channels are actually feeding AI answers, and which ones are just... there?

The problem is structural. A channel mix report built on GA4 or Adobe Analytics measures referred sessions. AI citation is a different event entirely — it happens when a model decides your page, your Reddit thread, or your YouTube video is worth quoting in an answer, whether or not anyone ever clicks through. You can have zero AI-referred sessions in your analytics and still be the most-cited source on a topic. You can also have decent AI traffic and be invisible on the citation side, quietly riding coattails of a page that ranks for reasons that have nothing to do with your GEO efforts.

So a real channel mix scorecard for AI citations needs to separate three things that get conflated constantly:

1. Citation performance — is this channel/content type actually getting quoted by AI engines?
2. Crawler access — are AI bots even reaching this content in the first place?
3. Referral impact — is this channel sending real, measurable traffic once you fix attribution?

Most scorecards stop at number three, and number three is the layer analytics tools get wrong most often.

Layer 1: score channels by citation share, not clicks

Start by pulling citation data per content type and per source domain, broken down by AI engine, because the picture changes depending on which model you're looking at.

The content-type shift is the headline finding right now. In July 2026, product pages became the single most-cited content format on ChatGPT at 32.8% of citations, nearly double their March share of about 18%. Listicles sit at 9.7% and growing fastest within the month, followed by news (5.2%), how-to content (4.1%), social posts (4.1%), and comparisons (2.9%). Video barely registers at 0.1%. Google AI Overviews moved the same direction in July: product pages overtook listicles as the top-cited format late in the month too. That's a genuine trend, not noise, and it means a scorecard that scores "blog content" as one undifferentiated bucket is missing the biggest swing in the data.

On the domain side, June 2026 data shows Reddit leading ChatGPT citations at 3.71%, more than six times the runner-up (Wikipedia at 0.56%), with no single domain clearing 4% — AI search is far less winner-take-all than classic Google rankings. But don't build a scorecard row that says "Reddit is safe." Reddit's ChatGPT citation share had already dropped from 6.11% in May to 3.71% in June, a roughly 40% single-month decline, and then it fell off a cliff entirely: from about 4% to 0.5% on August 14, 2026, tied to a platform access change. This is the second time it's happened — a similar collapse hit in September 2025, dropping Reddit from 10-14% of citations to roughly 1% within days, and it never fully recovered in that window.

If you're scoring channel performance today, weight that history into your risk column. A channel earning you 20% of your citations through one third-party platform isn't a strength, it's a concentration risk. You can pull and chart this kind of citation-share history for your own category with a tool like Promptwatch, which tracks the same domain-level citation trends across ChatGPT, Perplexity, Claude, Gemini, and Google's AI surfaces.

Promptwatch

Promptwatch

Track and optimize your brand visibility in AI search engines

View more

Layer 2: score by social platform, per engine, not in aggregate

Here's a mistake I see constantly: teams build one "social media" row in their scorecard and score it as a single channel. That collapses information you need.

Blended across engines, the social citation pecking order looks like Reddit (3.36%), YouTube (2.94%), Facebook (1.15%), LinkedIn (0.74%), Instagram (0.65%), TikTok (0.13%), and X (0.07%). But blend that data and you lose the actual pattern, which is that each engine has a completely different social diet.

ChatGPT is, functionally, a Reddit-first engine — 5.19% of its citations go to Reddit versus 0.23% for its next-best social source (LinkedIn). AI Overviews and Grok both prefer YouTube instead (4.08% and 4.85% respectively), and Perplexity leans YouTube too, at 2.67%. Grok is the outlier that actually gives meaningful weight to Facebook (3.01%) and Instagram (1.7%). X never breaks 0.25% share on any engine tracked, so if your scorecard has a row scoring X/Twitter effort against AI citation ROI, that row should be near the bottom every time.

The YouTube data has a specific practical lesson buried in it too. Cited videos skew mid-tier, not viral: the 10K-100K view bucket is the largest citation group, about 80% of citations go to videos under 100K views, and only 7% come from videos over 1M views. Likes barely matter, 43% of cited videos had under 100 likes. Same story for channel size, 10K-100K subscriber channels are the biggest citation bucket, 80% of citations come from channels under 100K subscribers. If your video strategy is chasing viral reach for AI visibility, that's the wrong target. Depth and topical relevance beat scale.

Reddit skews even more counterintuitive: about 71% of cited posts had fewer than 10 upvotes, and posts with 500+ upvotes made up just 0.6% of citations. Cited posts also skew old, roughly 60% were over six months old when cited, and posts under a month old made up only about 11%. Current AI visibility on Reddit reflects a historical community footprint you built (or didn't) a long time ago, not this week's activity. See Promptwatch's Reddit citation data from January 2026 and its YouTube citation report from February 2026 for the full breakdowns.

Layer 3: score by LinkedIn page type, because "LinkedIn" isn't one channel either

Same lesson, different platform. Aggregate LinkedIn citations across engines and Pulse articles look dominant at 37.67%, followed by posts (32.19%) and company pages (13.35%). But the per-engine split matters a lot for where you should be investing effort.

ChatGPT treats LinkedIn almost like a directory rather than a publishing platform, favoring company pages (23.84%) and even the bare LinkedIn homepage (22.55%) over Pulse articles (8.77%); job listings punch above weight there too at 8.02%. Google AI Mode is the opposite, hungry for actual articles: Pulse at 44.85%, posts at 27.37%. Perplexity flips the post-versus-article ranking entirely, with feed posts (41.88%) beating Pulse articles (32.46%). If your scorecard scores "LinkedIn thought leadership" as one line item, you're averaging together strategies that succeed on completely different engines for completely different reasons.

Layer 4: score crawler access, because citations require being read first

A content type or channel can score well on paper and still get zero citations if AI crawlers can't actually reach it. This is the layer most channel mix scorecards skip entirely, because it doesn't live in marketing analytics, it lives in server logs.

Crawler mix has been shifting fast in 2026. OpenAI's crawlers accounted for 79.8% of verified AI crawler requests in the week of August 31-September 6, 2026, down from 94.8% just a few months earlier in the week of June 8-14, a real shift toward Anthropic, Google, Perplexity, and Mistral crawlers taking a bigger share. Separately, Meta-WebIndexer jumped from roughly 2% to nearly 38% of all tracked AI crawler requests between mid-July and August 9, 2026, evidence Meta is building its own web index and a crawler most teams aren't yet checking their robots.txt or CDN rules against. If your logs only ever mention GPTBot, add a crawler-mix row to the scorecard now.

Building the actual scorecard: a template

Here's a workable structure. Score each channel from 1-5 on each dimension, then weight by business priority.

The "referral traffic (corrected)" column deserves its own explanation, because getting that number wrong undermines the whole exercise.

Fixing the referral layer before you trust it

AI referral attribution in standard analytics is genuinely broken in ways that are easy to miss. ChatGPT auto-appends utm_source=chatgpt.com to outbound links, but Gemini and Perplexity generally don't add UTM parameters, so a rules-based classification that only catches UTM-tagged traffic will miss most non-ChatGPT AI referrals entirely. Worse, in GA4, if a utm_source shows up without a matching referring-domain rule, the session often gets bucketed into "Unassigned" rather than any AI channel, which is why a lot of AI-referred traffic hides inside "Direct" or "Unassigned" rows without anyone noticing. Links opened with rel="noreferrer" or blocked by strict referrer policies strip the referrer header entirely too, pushing AI sessions into Direct traffic even with no UTM problem at all. Adobe Analytics has its own version of this: it doesn't process UTM parameters by default, so the same visit GA4 might catch lands as plain Direct traffic in Adobe, and Adobe's channel rules run top-to-bottom and aren't retroactive, so fixing the rule doesn't fix your historical data.

The practical fix is a custom channel grouping in GA4 or Adobe that ORs together UTM-source checks with referral-domain matches, rather than trusting the default channel definitions. And even after that fix, you still won't see which prompt drove the visit, because AI engines generally don't pass the original user query to the destination site. That gap is exactly why a citation-monitoring layer matters, tools like Promptwatch pair prompt-level citation tracking with visitor analytics and AI crawler logs so you can connect a specific citation to actual site visits, instead of guessing from analytics alone.

Where to get the data for each layer

You don't need five different subscriptions to build this, but you do need at least one tool that covers citation and crawler data, since neither shows up in standard marketing analytics.

For the citation and crawler layers specifically, Promptwatch tracks citation trends across ChatGPT, Gemini, Claude, Perplexity, Grok, and Google's AI surfaces, classifies citations into content types and source types, and logs which AI crawlers actually hit your pages and where they error out, which is the missing half of most channel mix conversations.

If you want to browse more options in this category before committing to one, the GEO software directory at bestgeosoftware.com lists the full range of AI visibility platforms, and the rank-tracking-focused ones are collected separately at ai-rank-tools.com.

Putting the scorecard to work

Once the scorecard exists, review it monthly, not quarterly. Citation shares move fast, the Reddit collapse in August 2026 happened in a single day, and a scorecard reviewed once a quarter would have missed the entire event until it was long over. Score trend direction as heavily as absolute share. A channel at 5% citation share trending down is a worse bet than a channel at 2% trending up.

And resist the urge to consolidate rows for simplicity. "Social media" as one row hides the fact that Reddit, YouTube, and LinkedIn behave completely differently across ChatGPT, AI Overviews, Perplexity, and Grok. The whole value of this exercise is in the disaggregation. Collapse it back down and you're just building a prettier version of the same channel report that already wasn't answering the AI citation question in the first place.