Best practices
How to actually get value out of Lumear. Lessons from the teams who treat AI visibility as a real channel rather than a quarterly check-in.
Get the matcher right before anything else
Lumear’s entire pipeline hangs off whether the matcher correctly identifies your brand in AI replies. Spend 10 minutes building a complete alias list before you start running prompts. A missing alias will make your visibility score look 30+ points lower than reality and every downstream insight will be misleading.
Run prompts on a real schedule
AI answers change month to month — sometimes week to week. A single run is a snapshot; trends are what matter. Most customers settle on:
- Weekly for the “monitoring” prompt set (25–50 prompts). Catches drift, sentiment swings, new competitor surfacing.
- Monthly for the “deep audit” prompt set (100+ prompts). Feeds your monthly visibility report.
- Ad-hoc after a content push. Rerun the affected prompts to measure whether the new pages moved AI behavior.
Design prompts the way buyers actually phrase them
The most common mistake is loading the prompt set with internal-marketing language. AI only echoes what real buyers ask. Use the prompt generator with a buyer persona, then read every prompt before saving. Reject anything that reads like a press release.
Three quick rules of thumb:
- Short and conversational beats long and clinical.
- If the question contains your brand name verbatim, the AI was always going to mention you — that’s less interesting than category-level prompts.
- Include 3–5 directly branded prompts per set anyway, for sentiment monitoring.
Mix branded and unbranded prompts
The default Mixed branding mode (~25% branded, ~75% unbranded) is the right starting point for most brands. Drop to Branded only when you want to track direct sentiment about your brand. Switch to Unbranded only when you want to measure organic surface area — “does AI bring me up unprompted?”
Watch the gap between mention rate and recommend rate
Mention rate alone is vanity. Recommend rate is the buying signal. If you’re mentioned 60% of the time but recommended 8%, AI knows the brand but is actively favoring rivals — that’s a content + messaging problem, not a discoverability one.
The Mentioned, not recommended insight surfaces this gap and links you straight to the worst-offending prompts. Treat those as the highest-leverage queue.
Don’t neglect citation diversity
Owned-domain citations are the most defensible signal — the AI is grounding its claim in content you control. If your citation mix is 70% Reddit and 5% owned, you’re one sentiment shift on a subreddit away from a problem. Editorial coverage and your own long-form pages are the levers.
Use Recommendations as a content queue, not a one-time list
Re-generate after every major content shipping cycle. Recommendations are computed from the most recent runs + the most recent crawl; both change as you work.