Studio openStart a project →
Level44

How we measured it.

This is the full method behind The State of AI Visibility for Northern Irish Business 2026. We’ve written it plainly so you can judge the findings, repeat the test, or cite it with confidence.

Scan date: 15 June 2026.

36 businesses, three sectors, twelve in each.

36 Northern Irish businesses across three sectors, twelve in each.

  • Sector 01

    Creative & web agencies

    A market that mostly relies on its own websites.

  • Sector 02

    Accountancy firms

    A market with active professional directories.

  • Sector 03

    Hotels & restaurants

    A market heavy with public reviews.

We chose sectors with different relationships to the open web. One heavy with public reviews (hospitality), one with professional directories (accountancy), one that mostly relies on its own websites (agencies). That spread lets us see whether footprint, not just quality, drives recommendation.

Four assistants. The same ones buyers ask.

Each business was tested against the four most-used assistants for open-ended recommendation questions in 2026.

  • ChatGPT

  • Claude

  • Gemini

  • Perplexity

9 questions per business, in two groups.

9 questions per business, designed to mirror how real customers ask. They fall into two groups.

Cold questions

Business name not in the prompt. The real test of discovery.

  • Category

    "Best [sector] in Belfast / Northern Ireland."

  • Problem-led

    "Who can solve [a specific problem]."

  • Recommendation

    "Recommend me a good [sector]."

Named questions

Business named in the prompt. A test of knowledge, not discovery.

  • Comparative

    "Is [business A] or [business B] better."

  • Direct

    "Tell me about [business]."

That’s 36 businesses × 9 questions × 4 assistants = 1,296 answers, all captured on 15 June 2026.

Six fields per answer.

For every one of the 1,296 answers, we logged the same six fields.

  • Presence

    Did the business appear?

  • Position / rank

    Where in the list?

  • Sentiment

    How was it described (positive, neutral, negative)?

  • Accuracy

    Were the claims about it correct?

  • Sources cited

    Every URL the assistant drew on.

  • Competitors named

    Which other businesses appeared instead.

Machine-tagged, human-checked.

We extracted the factual claims each assistant made about a business and compared them to ground truth. Claims were tagged provisionally by the system, then reviewed by a person, with 100% of flagged claims plus a sample of the rest checked. The accuracy figure reflects human checking, not just automated matching.

What counts as “cold visibility”.

Our headline visibility figures use only the three cold question types, where the business is not named, because that’s what discovery actually means. Including named questions, where the assistant is told who to talk about, inflates the number and measures recall, not discovery. We report both, clearly separated.

Stated plainly.

  • It's a snapshot, not a census.

    36 businesses on one day. It shows the shape of the problem, not a league table of every Northern Irish firm.

  • Assistants change.

    Models and sources update. A re-run on another day would shift the exact percentages, though we'd expect the pattern to hold.

  • Sector selection is deliberate, not random.

    We picked contrasting sectors on purpose. Don't read the three as representative of all Northern Irish business.

  • Cold-question framing matters.

    Different phrasings yield different rates. We standardised ours and report the question types separately.

Run on our own platform.

The scan was run on Level44’s own AI visibility platform, which queries the major assistants and logs presence, position, sentiment, accuracy and sources. It’s the same engine behind the free visibility check.

Cite this work

Level44 (2026). The State of AI Visibility for Northern Irish Business 2026. Methodology.

Published June 2026. Last updated: 17 June 2026.

About the author

Phill Hendry runs Level44, a Belfast brand and web studio building for the answer-engine era. This research was produced in-house using Level44’s own AI visibility platform.

Want to see how you score on the same method?

Run your domain through the same engine that powered this scan. Free, a couple of minutes, full PDF in your inbox.

Check your AI visibility. Free.

The series

Part of Level44’s AI visibility research set.

A connected set of pages on how AI assistants decide which businesses to recommend. Read in any order.