Skip to content

ALIQ AI Score: how AI assistants treat a business

ALIQ AI Score is ALIQ Studio's in-house measure of how AI assistants treat a business. Every week we ask ChatGPT the same 48 buyer questions, three times each, and grade every answer for whether it mentions, recommends and cites the business, how it describes it, and whether it gets the facts right.

We built it to measure ourselves first, and it works for any business: on the AI visibility retainer, every client gets their own ALIQ AI Score in their monthly report, worked out on their own buyers' questions. Most agencies that sell AI visibility describe a tool and show a score, but not how the score is made. This page is how ours is made.

What each answer is graded on

Five things, kept separate because they are different results. An assistant can name a business without recommending it, and cite a page without naming who wrote it.

Mentioned
Does the answer name the business, or show its website as a source?
Recommended
Does it tell the person asking to use the business?
Cited
Does it link to the business's own website as a source? This one is decided by matching the website address, not by judgement.
Sentiment
Does it describe the business favourably, neutrally or unfavourably?
Accurate
When it names the business, does it get the facts right? Answers are checked against facts the business has published, such as its prices.

How the score works

The score runs from 0 to 100. Being recommended counts for 40 points, being cited for 25, sentiment for 20 and accuracy for 15. Each part is the share of all answers in which it happened. A favourable answer counts in full, a neutral one counts half, and an unfavourable one counts nothing.

Because every part is counted across all the answers, a business that no answer mentions scores zero. It cannot score well just because nobody says anything bad about it.

A worked example

Say 100 answers are graded. 10 recommend the business, 10 cite its website, 12 mention it favourably and 4 neutrally, and the 16 that mention it all get its facts right.

Recommended
10 of 100 answers, × 40 = 4.0
Cited
10 of 100, × 25 = 2.5
Sentiment
12 favourable + 4 neutral at half = 14 of 100, × 20 = 2.8
Accurate
16 of 100, × 15 = 2.4
Score
4.0 + 2.5 + 2.8 + 2.4 = 11.7 out of 100

The questions

48 questions a buyer might ask about marketing in Malaysia, 42 in English and 6 in Bahasa Malaysia, across seven services: websites, SEO, social media, content, advertising, branding, AI search visibility. They do not change from one week to the next, so one week can be compared with the last.

Each question is asked three times, because an AI assistant does not give the same answer every time. One answer is a single draw. Three show whether a result holds, and the result that holds is the one we report, not the best one we saw.

We do not publish the questions themselves. A fixed set that anyone can read is a set that can be written for.

Comparing agencies the same way

On the same questions we also grade the agencies AI assistants name most often, with the same rules. For that comparison we use a Visibility score: the ALIQ AI Score without the accuracy part, so out of 85 and scaled to 100. Accuracy needs a business's own facts: we hold ours, and a client's when they share them with us. For the businesses named alongside them we have no facts to check against, so the comparison leaves accuracy out.

What the score cannot tell you

  • AI answers change with the model, the account, the location and the day. A score describes the week it was measured.
  • A mention is not a sale. Being named by an assistant is a step towards an enquiry, not the enquiry itself.
  • Nobody controls what an AI assistant says, so nobody can guarantee that one will recommend a business, us included.

Get your ALIQ AI Score

Your first AI visibility measurement is free. It shows which questions name you, which name your competitors instead, and which websites the answers cite. On the retainer, your monthly report includes your ALIQ AI Score.

See how AI visibility work runs

Method last updated: