To check whether ChatGPT or another AI assistant recommends your Irish business, you need to ask it a series of realistic customer queries and record exactly how it names you, describes you, and whether it cites a source. A single positive mention is not a reliable signal; you need a consistent pattern across multiple questions and multiple sessions. This article gives you a practical framework to do that systematically.

Not every mention by an AI assistant carries the same weight. There are broadly three tiers you will encounter.

Confident direct mention: The assistant names your business specifically, gives your location, and may include a detail such as your opening hours, a service you offer, or a review score. This is the strongest signal and the one worth tracking most carefully.

Vague category suggestion: The assistant says something like "there are several good plumbers in Cork — you could check Google Maps or Golden Pages." Golden Pages is Ireland's long-running business directory, equivalent to the old printed yellow pages. This kind of answer acknowledges your category but does not name you; it is not a recommendation in any meaningful sense.

Hedged mention: The assistant names you but immediately qualifies it — "I'm not sure if they're still open" or "you should verify this." This is a partial signal. It shows your business exists in the model's training data, but the AI does not have enough confidence to recommend you without reservation.

For measurement purposes, only the first tier counts as a genuine recommendation. The other two tell you something useful — you are in the right category, or you have some presence — but they should not be recorded as wins.

A measurement framework: questions you test, the business attributes you expect to be repeated, and acceptance thresholds

A measurement framework here means a fixed set of test queries, a list of attributes you expect the AI to get right, and a clear rule for what counts as a pass.

Step 1 — Build your query set. Write five to ten questions a real customer might type or speak. For a Dublin accountant, that might include "who does self-assessment tax returns in Dublin 6," "best accountant for a small business in Rathmines," and "can you recommend a local accountant near Ranelagh." Vary the phrasing; AI assistants are sensitive to wording.

Step 2 — List your expected attributes. For each query, decide which facts the AI should reproduce if it genuinely knows your business: your trading name, your town or neighbourhood, your main service, and ideally one differentiator such as "open Saturdays" or "Irish-speaking staff."

Step 3 — Set an acceptance threshold. A reasonable threshold is that your business is named in response to at least half your queries, with at least two correct attributes each time. Below that, your AI visibility is weak. Above it, you have a baseline worth protecting and improving.

Run this test across at least two different AI platforms — ChatGPT and Google's Gemini are the most used in Ireland currently — because their training data and update cycles differ.

Margin and repetition: why your score may change even when nothing on your site seems to change

AI language models (large-scale systems trained on text from across the internet) are retrained or updated on schedules you do not control. When a new version rolls out, the weighting given to different sources can shift. A business that was mentioned confidently in one version may be mentioned more cautiously in the next, or not at all.

There is also natural variation within a single model. Ask the same question twice in the same session and you may get a slightly different answer. This is called stochasticity — the model does not always produce identical outputs. It means a single test result is not reliable; you need to repeat each query at least three times and look at the pattern.

Third-party platforms compound this further. If Tripadvisor updates your review count, or if a local news article that mentioned you is removed, the AI's next training run may weight your business differently. You did nothing; the landscape around you changed.

This is why a monthly tracking cadence is more useful than a one-off check.

APIs vs apps for tracking: what to automate, what to review manually, and where false confidence happens

An API (Application Programming Interface) is a way to query an AI model programmatically — sending questions and receiving answers automatically, without opening a chat window. This is useful for volume: you can test dozens of queries across multiple models in minutes.

However, automation introduces a risk of false confidence. APIs often return responses in a slightly different format than the consumer app, and some safety filters or response styles differ between the two. A business that appears in an API response may be described differently in the app a customer actually uses. Always manually verify a sample of your results in the real consumer interface — the ChatGPT website, the Gemini app, or the Bing Copilot sidebar — before drawing conclusions.

What to automate: running your fixed query set on a regular schedule and logging whether your business name appears.

What to review manually: the quality of the mention (confident vs hedged), the attributes cited, and whether a source is named. When an AI cites a source — a Tripadvisor listing, a Google Business Profile, or an article on a local site like the Irish Examiner — that citation tells you which piece of content is doing the work.

A Google Business Profile (GBP) is a free listing on Google that shows your address, hours, phone number, and reviews in Search and Maps; it is one of the most commonly cited sources in AI-generated local answers.

An evidence-first improvement loop: adjust the source, test the query, confirm AI citations, then iterate

Guessing what to change and hoping for a better score is slow and unreliable. An evidence-first loop is more efficient.

  1. Identify the gap. Your query set shows the AI does not mention your business for "beauty salon Galway city centre." That is the gap.
  2. Find the likely source. Check which platforms a competitor who is mentioned appears on. Are they on Booksy (an online booking platform popular with salons and barbershops in Ireland)? Do they have more Google reviews? Is there a local directory listing you are missing?
  3. Make one change. Add or update a single source — claim your Booksy profile, add your address and category to Golden Pages, or update your GBP with current hours.
  4. Wait and retest. AI models do not update in real time. Allow four to six weeks, then run your query set again.
  5. Confirm the citation. If your score improves, check whether the AI now cites a source. If it does, you know which platform is doing the work. If it does not cite a source but mentions you, the change may have reinforced something already in the training data.
  6. Iterate. Move to the next gap in your query set.

This loop is slow by the standards of social media marketing, but it is the only honest way to know what is actually working.

Week one is for establishing your baseline. Run your full query set across two AI platforms, record every result, and note which tier each mention falls into. Do not make any changes to your listings yet; you need a clean starting point.

Week two is for identifying your highest-priority gap — the query type where you are most absent — and making one targeted change to a third-party source.

Weeks three and four are for monitoring. Re-run a subset of your queries every few days to catch any early movement, but do not over-interpret short-term noise. The meaningful signal is the comparison between your week-one baseline and your end-of-month results.

If you want a quick starting point before building out your own tracking sheet, the free check on suggestedby.ai measures whether the main AI assistants name your business — it gives you a concrete baseline in a few minutes rather than hours.

Reading trends over time means looking at direction, not just absolute scores. A business whose score is gradually rising over three months is in a healthier position than one with a high score that has been flat or declining. Trends tell you whether your efforts are compounding or whether you need to revisit your source strategy.

The core discipline is consistency: same queries, same platforms, same recording method, every month. That is what turns a vague sense of "am I visible to AI?" into something you can actually manage.