Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- AI Search
Would you put share of voice in AI answers into a board report yet, with no platform offering query-level clicks? What would make it credible?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
Yes, I would put it in a board report now, but as a clearly labelled context metric, not as a performance KPI and not next to revenue as if it were comparable. Boards are already hearing about AI search from other directions. If SEO says nothing, someone else fills the gap with a vendor screenshot. The credibility problem is not the lack of query-level clicks as such; it is that most AI share-of-voice numbers are presented with more precision than the method can support. Fix the method and the presentation, and it earns a line.
What you can actually measure today
It helps to be clear, in the report itself, about what each platform does and does not tell you.
Source What it gives you What it does not Search Console “Search generative AI performance” reports Impressions for AI Overviews, AI Mode and generative AI features in Discover, by page, country, device and date Clicks, CTR or queries Bing Webmaster Tools AI Performance report Citation counts, cited pages and a sample of grounding queries across Copilot and Bing AI summaries Clicks or traffic GA4 “AI Assistant” channel Sessions from recognised assistants such as ChatGPT, Gemini and Copilot Traffic arriving without a referrer, and Google’s AI Overviews and AI Mode, which stay in organic search Merchant Center AI performance insights Share of voice for product categories in English-language queries in a handful of markets Anything outside retail, or product-level detail Prompt-tracking tools How often your brand appears in answers to a fixed prompt set, compared with competitors What real users actually asked, or how often Only the last row is a share-of-voice measure in the sense the board will understand, and it is a sample you design. That is fine as long as the report says so.
Why most AI share-of-voice numbers are fragile
Answers vary from run to run. The SparkToro and Gumshoe study from late 2025, which had volunteers run the same prompts through several assistants thousands of times, found that the exact list of brands recommended almost never repeated, while how often a given brand appeared across many runs was far steadier. The practical conclusion is the one the authors drew: appearance rate across many prompts and runs is a usable metric; an AI “ranking position” is not.
Three other things undermine credibility if you ignore them:
- Prompt selection. A prompt set written by the marketing team tends to flatter the brand. If the prompts are chosen after you see the results, the metric is worthless.
- Model changes. A model upgrade can move every brand’s share overnight. Without annotations, the board will read that as your performance.
- Blending platforms. ChatGPT, Google AI Mode, Perplexity and Copilot retrieve and cite differently. A single combined percentage hides the one platform that actually matters for your buyers.
What makes the number credible
Fix a prompt set tied to revenue categories before measuring, and version it↓Run each prompt many times per platform, on a set schedule↓Report appearance rate with a range, per platform, against named competitors↓Annotate model updates, prompt-set changes and major launches↓Show it next to outcome metrics it should lead, not instead of themA few specifics make the difference in the boardroom:
- Weight the prompt set by commercial value, not volume. Group prompts by product line or buyer problem, and show share per group. “We appear in about half of answers about payroll software for small firms, and rarely in answers about enterprise payroll” is a decision-grade sentence. A single blended number is not.
- Show uncertainty. If appearance rate is 42% across 30 runs of 50 prompts, show a range or at least say that movements of a few points are noise. Boards respect a stated margin more than false precision.
- Keep the method stable for a year. Add new prompts as a separate set rather than rewriting the old one, so trends stay comparable.
- Pair it with outcomes. Put AI Assistant sessions, branded search demand (the Search Console branded queries filter helps here), direct traffic and pipeline or revenue from those sources beside it. If share rises and none of those move over two or three quarters, that is useful information too.
- Say what action it drives. A metric earns a place in a board pack when it informs a decision: which product lines need third-party coverage, where competitors are winning recommendations, or whether a PR programme is changing how assistants describe you.
Hypothetical example of a board line:
“AI answer visibility (sampled): across 60 buying-intent prompts in our three core categories, run 20 times each on ChatGPT and Google AI Mode this quarter, we appeared in 38% to 44% of answers, up from 30% to 35% last quarter. Our main competitor appears in about 60%. The rise follows our analyst report coverage; a model update on ChatGPT in the same month makes the size of the change uncertain. AI assistant referrals were up 22% and represent 1.5% of site sessions. Next quarter we are targeting the enterprise category, where we rarely appear.”
That is a context line with a method note, a competitor comparison, an outcome metric and an action. It does not claim revenue.
What I would leave out
- Any figure presented as “AI traffic value” or “AI revenue” built from impressions. Search Console’s AI report has no clicks, so any revenue number derived from it is an assumption stacked on an assumption.
- Vendor “visibility scores” whose method you cannot explain in one sentence to a non-specialist.
- Month-on-month changes smaller than your run-to-run noise.
- A combined “AI share of voice” across every assistant without platform breakdown.
How to position it with the board
I would present AI visibility as a leading indicator of brand consideration, in the same family as share of search or unaided awareness, and explicitly not as a channel with attributable revenue. That framing is honest, it matches how the data behaves, and it protects you when the number moves for reasons outside your control. If a director asks why there are no clicks, the answer is simple: none of the platforms provide query-level click data for AI answers yet, so the report uses the measures that exist and says where they stop.
The test of whether it belongs in the pack is whether it survives a hard question. If you can explain how prompts were chosen, how often they were run, what a normal fluctuation looks like and what you will do differently because of the number, it is credible enough. If you cannot, keep it in the marketing appendix until you can.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.