Why this does not fit inside one team
The sources that shape an answer are spread across the company. Documentation belongs to product. Reviews are influenced by support and customer success. Coverage belongs to PR. Integration listings belong to partnerships. The site belongs to marketing. No single team can ship all the fixes a baseline turns up.
The most useful single artefact for this is a map: each kind of source that shapes your answers, and the team that owns it. It turns an abstract channel into a list of names.
Getting leadership to back it
The case is strongest when it is concrete. Show real answers, not a model of them:
- What the engines say about you today, with the answers themselves.
- The sentence you want them to say, and the distance between the two.
- The plan, broken down by the team that owns each piece.
- How progress will be measured, and when the first read is due.
- What you need from each team, stated as a specific ask.
The goal is a shared objective across the teams involved, so the work is part of their plans rather than a favour.
What to measure
Measure in three layers. They move at different speeds, and each is honest about something the others cannot see.
| Layer | What to track | How it behaves |
|---|---|---|
| The answer | How often you are named, share of voice against competitors, sentiment, factual accuracy — per engine | Moves first, and is the most direct evidence of the work |
| Visits | Referrals from assistants, AI crawler fetches of key pages, branded and direct traffic | Partial: most influence produces no click |
| Revenue | Pipeline where buyers say they found or chose you through an assistant | Slowest, and self-reported rather than modelled |
Resist the pressure to turn the first layer into a modelled revenue number. Report the answer layer as the leading indicator, the visits as confirmation, and revenue only where buyers themselves said it.
Two levels of goal
| Programme goals | Initiative goals |
|---|---|
| How often you are named across the whole prompt set, per engine | How often you are named on the prompts for one claim |
| Share of voice against your main competitors | Sentiment on one recurring objection |
| Visits referred by assistants | Whether one wrong fact has stopped appearing |
| Pipeline buyers attribute to assistants | Whether a target page is being fetched and cited |
Tie initiatives to the answer, not to revenue. A single page or correction is too small to show up in pipeline, and pretending otherwise discredits the numbers that are real.
Who runs it
One team should own the programme. It is often SEO or content, because they already think in pages and questions. That team keeps the prompt set, runs the measurement, sets priorities and writes the briefs.
The other teams do the work that sits with their sources, and it goes further when it is tied to goals they already have. Support wants fewer repeat questions, which clearer documentation also delivers. PR wants coverage, which is more valuable when it lands on pages the engines cite.
A rhythm that holds
- Weekly: a short look at movement, and a check on anything that just shipped.
- Monthly: a review claim by claim with the owners of each piece of work.
- Quarterly: revisit the positioning sentence, the claim list and the prompt set, with any changes made in a dated batch.
Where to start tomorrow
- Write the sentence you want an engine to say about you.
- Pick the three claims that matter most to buyers.
- Write a handful of buyer-worded prompts for each, and ask them of every engine you care about.
- Read the answers and the sources, and write the three lists.
- Take one finding to the team that owns it.
Key takeaways
- Map each source that shapes your answers to the team that owns it.
- Win backing with real answers, a clear gap, and specific asks per team.
- Measure the answer first, visits second, and revenue only where buyers say it.
- One team owns the programme; others do the work tied to goals they already have.