You don't need a programme to start. You need a baseline, one hypothesis and a date to look again. Here's a four-week plan that fits next to a normal job.
#Week 1 — Baseline
Goal: know where you stand, with evidence.
- Pick one business area (a product line or audience), not the whole company.
- Draft 20–30 buyer questions across problem, shortlist, comparison and constraint stages. Our prompt builder is a decent start; swap in real questions from sales calls.
- Run them on the engines your buyers use. Save the answers, the cited sources and the date.
- Record three numbers separately: mention rate, owned-citation rate and recommendation rate.
Output: a one-page baseline with the panel, the numbers and three example answers.
#Week 2 — Investigate
Goal: separate access problems from content gaps.
- Run the AI crawler checker on your domain. Are search and user-fetch crawlers admitted?
- Run the page-readiness audit on your five most important pages. Is the key claim in the raw HTML?
- List the top cited domains across your answers. Which cite your competitors but not you?
- Read five answers that omit you. What do they say instead?
Output: a ranked list of causes — access, page, source — with the evidence for each.
#Week 3 — Ship one improvement
Goal: make one bounded change you can attribute.
Choose the cause with the best ratio of impact to effort. Examples:
- Fix an accidental robots.txt block.
- Rewrite one comparison page so its opening paragraph states the answer, the audience and the key limit.
- Add the missing specifics to documentation — supported integrations, limits, prerequisites — instead of adjectives.
Resist the urge to do three things at once. If you change the page, the panel and the schedule together, you won't know which mattered.
Output: the change, its date and the prompts it should affect.
#Week 4 — Re-measure and write it up
Goal: read the result honestly.
- Rerun the same panel under the same conditions, several times.
- Compare against the baseline range, not a single earlier run.
- Check the specific prompts the change targeted, then the panel overall.
- Write down what happened — including if nothing did.
A flat result is information. It might mean the change was too small, the effect hasn't propagated, or the cause was elsewhere. Don't promise a timeline you can't back with data.
#After the first month
Pick the next hypothesis from the ranked list and repeat. Over a quarter you'll have a record of what you tried, what moved and what didn't — which is more valuable than any single score.