In-House AI Visibility Tracking: Prompts, Evidence and Weekly Review

Question 1

How can a small business automate brand monitoring in AI search in-house?

Accepted Answer

The article recommends automating collection with a prompt sheet, a scheduler, and API calls that check each answer for the brand name and write results back on schedule. With 25 prompts, three runs, and four weekly cycles, the article counts about 300 requests a month for one provider before other providers and reading time. The article states teams should keep one row per answer and store the full answer text, not just yes-or-no, so the cited URL and competitor page can be re-read later.

Question 2

Which prompts should I track for AI visibility?

Accepted Answer

The article recommends starting with 20 to 30 real buyer prompts written exactly as buyers type them, drawn from sales calls, support tickets, and buyer communities. According to groas, prompts should be grouped like ad groups into discovery, comparison, problem, brand, and local if location matters, with brand prompts scored separately because winning your own name can mask losses elsewhere. Wording of the core set should be frozen so week-to-week comparisons hold, with a small exploratory set rotated monthly and labeled separately.

Question 3

Why is checking each AI prompt once not enough to measure visibility?

Accepted Answer

According to groas, one run is a snapshot rather than a measurement because answers vary: in a large test with 600 volunteers running 12 prompts almost 3,000 times, fewer than 1 in 100 pairs produced the same brand list and fewer than 1 in 1,000 matched in the same order. The article states teams should keep three questions apart — whether the brand was mentioned or cited, whether people click through, and whether crawlers can fetch pages — because a rising mention rate can sit next to flat referral traffic for months. The report’s job is to show where the brand appears, disappears, and what answers point to, to choose an investigation.

Question 4

What should I look for in the weekly AI visibility review?

Accepted Answer

The article recommends sorting fixed prompts into Gained, Lost, and Never-had before looking at aggregate rates: Gained means appearing in at least two of three runs after missing last week, Lost means disappearing after previously holding, and Never-had means never appearing while the same competitor keeps winning. According to groas, noise looks like a single-run flip with no change in cited URLs, while movement worth investigating looks like a shift across two consecutive weeks with a new cited domain or a mention rate dropping from 80% to 20% on the same fixed wording. Two weeks plus a URL change means work, while one lonely flip should be ignored and the work should start with reading the answers.

Question 5

When should I stop tracking AI visibility in-house?

Accepted Answer

The article estimates a 25-prompt setup takes four to six hours to build the first time, then about 60 to 90 minutes a week to read the diff plus another hour a month for maintenance, or 6 to 8 hours a month in total. As an example, at $20k a month on search with time valued at $150 an hour, that is $900 to $1,200 in labor to keep the report honest. According to groas, teams should skip DIY entirely with more than 50 prompts, more than three locations with different prompt sets, or nobody who will open the sheet on Monday, and should start with 15 prompts covering five brand, five comparison, and five problem questions.