Practical Lens 60
Practical Lens 60: AI visibility needs competitor prompts
Testing "What does our company do?" measures recognition. Buyers more often ask who they should consider in a category — that is where you discover whether AI answers put you, a competitor or nobody on the shortlist.
Board-level signal
Brand visibility is not category visibility. A company can appear in answers about itself while being absent when a buyer asks for a recommended provider.
What this lens means
Competitive prompts are buyer-like questions that do not name your company. They test the market decision, not your existing awareness. A useful prompt includes the category, relevant situation, constraint and selection criterion. The output is observed evidence — not a market ranking or a claim that one AI answer represents every buyer.
Brand prompt
A question about a named company. Useful for factual accuracy, but weak for measuring category competition.
Category prompt
A question a buyer can ask before knowing which vendors exist. Useful for observing recommendation visibility.
Key terms
- Prompt set
- A fixed group of buyer-like questions run consistently so results can be compared over time.
- Recommendation presence
- The number of tested answers that name the company as a relevant option, divided by the number of valid tests.
- Competitor appearance
- A named alternative in an answer to the same category prompt. Record it; do not infer a market share from it.
- Test protocol
- The fixed platform, prompt wording, date, response capture and scoring rule used by every reviewer.
Why this happens
Companies naturally start with their own name, products and website. Buyers rarely start there. They ask for the best fit, a safe option, an alternative or a provider for a specific situation. If measurement follows internal language rather than buyer language, it confirms what the company already knows and misses the competitive decision point.
What this usually indicates
- AI monitoring is organised around the brand rather than the buyer journey.
- Marketing cannot show which competitors appear for the same category question.
- Prompts change each month, so results cannot be compared responsibly.
- Leadership sees a mention count without the selection context that gives it meaning.
What to verify (evidence-only)
- Collect real buyer questions. Use sales calls, lost-deal reasons, search queries and customer interviews. Remove brand names from the final test prompts.
- Create a compact prompt set. Start with 10–20 questions covering the highest-value use cases, geographies, constraints and decision criteria.
- Set a scoring rule before testing. Define what counts as mentioned, recommended, cited and not present. Do not change the rule after seeing an answer.
- Capture the answer. Record prompt, platform, date, full response, named companies, order of appearance and cited sources where visible.
- Review patterns, not isolated outputs. Calculate results only across the fixed set. Escalate a trend after repeated observations, not after one surprising answer.
Evidence log template
Track each competitor-prompt test as a comparable record. A category question with no register entry, or an entry with no review date, is the gap to close first.
prompt,our_company,competitors_named,recommendation_or_citation,review_date
[category + use case + constraint],present / absent,[names as observed],[exact observation],[date]
[category + use case + constraint],present / absent,[names as observed],[exact observation],[date]Terminal check: calculate recommendation presence?
Save a reviewed test table as prompt-results.csv with the columns prompt,valid,our_company_recommended. Use 1 for yes and 0 for no. The expected result is a transparent rate based only on valid tests.
awk -F, 'NR>1 && $2==1 {valid += 1; recommended += $3}
END {if (valid) printf "Recommendation presence: %.1f%% (%d/%d)\n", 100*recommended/valid, recommended, valid; else print "No valid tests"}' prompt-results.csvBusiness meaning: this is a repeatable observation rate for your chosen prompt set. It is not a universal market-share number and should always be read with the prompts and dates used.
PowerShell check: calculate the same rate?
Use the same reviewed CSV. The expected result is the number of valid category tests and the share that recommended your company.
$rows = Import-Csv .\prompt-results.csv | Where-Object { $_.valid -eq '1' }
$recommended = ($rows | Where-Object { $_.our_company_recommended -eq '1' }).Count
if ($rows.Count -gt 0) {
'Recommendation presence: {0:N1}% ({1}/{2})' -f (100 * $recommended / $rows.Count), $recommended, $rows.Count
} else { 'No valid tests' }Business meaning: retain the underlying answers. The rate is useful only when leadership can inspect what each observed recommendation meant and which buyer question produced it.
Decision rule for CXOs
Do not approve an AI visibility dashboard that reports only brand prompts. Require a fixed competitor-prompt set for the category decisions that create pipeline. The aim is not to chase every answer; it is to see the shortlist buyers may encounter before they know your name.
Frequently Asked Questions
Why are brand-name prompts not enough?
They test recognition after the buyer already knows the brand. They do not test the category questions where a buyer may first encounter alternatives.
What makes a competitor prompt useful?
It reflects a real buyer decision: category, geography, company size, use case, constraint and selection criterion — without naming the company being measured.
What should be recorded?
Record the exact prompt, platform, date, response, named companies, recommendation position, citations and whether required decision criteria were addressed.
Does one answer establish market position?
No. One answer is only an observation. A repeatable prompt set and a stable method are needed to identify patterns over time.
VerisAI, founded in 2025, helps B2B teams build evidence-led AI visibility monitoring around real buyer decisions.