AI citations: the share of voice metric for CMOs

September 22, 2026 · 5 min · Free, with full transcript

Comscore puts ChatGPT at 43% of AI-driven discovery, down from over 50%, with Gemini and Claude taking share. The question here is not who wins the assistant race but what happens to measurement when discovery, comparison and purchase collapse into one AI turn that leaves no click, no session and no referral trail. The position: last click attribution now credits brand or nothing, and soft organic traffic can hide influence you actually won. You leave knowing which dashboard lines mislead you, how to read vendor numbers from SEMrush and HubSpot against Forrester's independent view, why branded direct traffic and branded search are the cleanest remaining signal, and how to run a monthly citation audit across ChatGPT, Gemini and Claude.

Listen to the podcast

4 min

Chapters

Key takeaways

  • Add a "mentioned but not visited" view before blaming the algorithm for soft organic traffic.
  • Stop trusting last click attribution when the last touch is a direct visit or branded search after an AI recommendation.
  • Treat SEMrush citation gap figures and HubSpot's one third AI influenced pipeline as vendor best case, and cross check with Forrester.
  • Track branded direct traffic and branded search spikes as your new referral log for AI recommendations.
  • Pick your 10 highest value buyer questions, ask ChatGPT, Gemini and Claude each, and record whether you are named, how and against whom.
Read the full transcript

Host:This is Leaders Insights. On the table, AI citations are becoming the new share of voice metric. 43%. That's ChatGPT's share of AI-driven discovery this week. Down from a number that used to have a 5 in front of it, ComScore just published it. Is this the story?

Expert:The share drop is the headline everyone's chasing. It's the boring part. Gemini and Claude eating into ChatGPT is a horse race and horse races end. What doesn't end is what the fragmentation does to your ability to measure anything at all. Walk

Host:me through that. Why does one assistant losing ground to two others break

Expert:measurement? Start with how discovery used to work. Someone had a problem. They searched. They clicked a few links. They compared. They bought. Every one of those steps left a footprint. A click. A session. A referral source you could trace back. Attribution, which is just working out which touchpoint gets credit for the sale, lived in those gaps between steps. And the AI interaction closes the gaps. It collapses them into one turn. Somebody asks Claude, what's the best project management tool for a 10-person agency? And Claude names three. No search page, no comparison clicks, no referral trail. The discovery, the comparison, and half the purchase decision happen inside a black box you can't instrument. So the

Host:metric that replaces it is what? This AI citation idea? How often an assistant

Expert:names you in what context and whether it links you. Think of it as share a voice for a world with no shelf. Share a voice. The slice of the conversation your brand owns versus rivals. Except the conversation is now a machine

Host:deciding whether you exist in its answer. Let's tear one apart. If I'm a CMO looking at my dashboard Monday morning, what's on it that's now lying to

Expert:me? Your organic traffic line first. It'll look soft and your team will blame the algorithm. Some of that traffic didn't leave. It never clicked through because the assistant answered on your behalf. The visit died. The influence didn't. Your dashboard has no column for mentioned but not visited. So it reports a loss where you actually won. What else? Your last click attribution, crediting whichever touchpoint came right before the sale. When someone buys after an AI recommendation, the last click is often a direct visit or a branded search. So the dashboard credits your brand or credits nothing and your paid team quietly takes a bow for a sale Claude actually made. That's uncomfortable.

Host:Whose budget is wrong then? Everyone's slightly. SEMrush has a tool now

Expert:that tracks how often you appear in AI answers and they'll tell you the citation gap between you and competitors is enormous. Worth remembering, SEMrush sells the software that measures it. So cross check the scale with someone who isn't invoicing you. Forrester's independent work says most marketing teams still can't see AI source demand at all. They're flying with an instrument that stopped moving. HubSpot's been loud about AI attribution being solvable inside their platform. Buy it? HubSpot puts AI influenced pipeline at something like a third of new deals for their customers. They sell the CRM that reports that figure. So treat it as a vendor's best case, not gospel. The direction is right. The precision is marketing. Nobody has clean attribution for a conversation that happened inside a model they don't

Host:own. So what actually works on that dashboard? Name the piece that

Expert:survives. Branded direct traffic and branded search. People typing your name in cold. That's the closest on a signal you've got. If an assistant recommended you, the human often goes and searches you directly to confirm. That spike is your new referral log. It's crude, but it's not lying. Give me the one thing to do Monday. Pick your 10 highest value buyer questions. The actual sentences a customer would type. Ask Chat GPT, Gemini and Claude, each one this week. And write down whether you're named, how, and against whom. That's your baseline citation audit. Do it monthly. It costs an afternoon and tells you more than the

Host:traffic report your board's been staring at. An afternoon and some honesty.

Expert:Cheaper than the software. Considerably. And it doesn't send you a renewal invoice.

Host:What we read for this one. Forrester Research DigiDay SEMrush vendor SEO analytics tools HubSpot vendor CRM marketing automation. We'll leave it there. The CMO self-assessment at MBA-training.com will tell you where you really stand.

Read and go deeper