Why we run every prompt multiple times, and why that's not a minor detail
Most AI-visibility audits ask a question once and report the answer as fact. AI outputs aren't consistent — that approach is statistically weak, and it's the single thing most publicly available audits get wrong.
Why this is harder to fake than it sounds
Running a prompt once is fast and cheap — which is exactly why most audits, from agencies and from free online tools alike, do it that way. Running it properly, multiple times per engine, at a price that still makes sense for a small business, only works if the process is automated rather than done by hand.
That's the actual reason the audit costs £295 and not £2,995 — the rigor comes from tooling, not from billable hours. We built it this way on purpose, not because it was the easy option.
The problem with asking once
A single-prompt audit
Asks ChatGPT one question, checks whether the brand appears, and reports 'cited' or 'not cited.' But ask the exact same question five minutes later and you might get a different answer entirely — AI outputs vary run to run, even with identical input.
Run 1: cited → report says 'you're visible'
Litealign's approach
Runs the same prompt multiple times per engine and reports a rate, not a verdict. A rate is honest about uncertainty. It tells you how reliably you show up, not just whether you happened to once.
5 runs: cited 2/5 → citation rate: 40%
How an audit actually runs
Four steps, the same for every client.
Build a prompt set specific to the client
Not generic questions — the actual phrases a real buyer would type into ChatGPT or Perplexity when evaluating this specific category or company.
Run each prompt multiple times, per engine
Typically 3–5 runs per prompt across ChatGPT, Gemini, Perplexity, and Google AI Overviews — because a single response can't be trusted to represent the norm.
Score citation rate, position, and sentiment
Not just 'mentioned or not' — where in the answer, how prominently, and whether the mention is favourable, all get tracked separately.
Benchmark against real competitors
The same prompt set, run against the brands actually being compared against — so the score means something relative, not just absolute.
Who built it
Master's in Artificial Intelligence
This methodology was designed and is maintained by Litealign's data & measurement lead — not adapted from a generic template.
See it applied to a real example
See the methodology in a sample report.
A full sample report, built exactly the way yours would be.