What Is AI Search (GEO) and Why Irish Businesses Can't Ignore It in 2026

The 30-second version

AI search means search engines that read your question and return a written answer instead of ten blue links. ChatGPT Search, Perplexity, Google's AI Overviews and Gemini all do this. They pick a handful of businesses to recommend per query. The discipline of getting cited is GEO (Generative Engine Optimization).

Why this matters now

• One in five Irish search queries now happens inside an AI tool.
• Google's AI Overviews now appear on roughly 30% of UK/IE commercial searches.
• AI search users convert at 2–3× the rate of traditional search users.

What AI search tools read

Proper H1 → H2 → H3 hierarchy. Short paragraphs (2–4 sentences). Clear factual statements. FAQ sections in question-and-answer format. A plain-text file at yourdomain.ie/llms.txt listing your key pages, services and a brief business summary. Takes 10 minutes.

• LocalBusiness with NAP, hours, service area.
• Service for each service.
• FAQPage for FAQ sections.
• Review if you display reviews.

Citation, not click

AI search is about being named in the answer. A user asks ChatGPT "best emergency plumber in Wexford"; ChatGPT lists three names; the user calls one. They never visit your site. New success metrics:

• AI Presence Rate — how often you're cited for queries you care about.
• Citation Authority — top of the answer vs. footnote.
• Phone calls and direct enquiries.

The 30-day GEO checklist

1. Add schema markup to every page. 2. Rewrite service pages in question-and-answer format. 3. Publish a /llms.txt file. 4. Three FAQ-rich pages targeting real customer questions. 5. Get cited in three local sources — chamber directory, LEO website, an Irish trade publication.

Is it working?

Search your own queries in ChatGPT, Perplexity and Google AI Overviews. Are you named? Watch monthly. Slow, then sudden — that's how AI citation builds.

How an answer engine decides who to name

An AI answer is assembled, not retrieved. The model runs its own searches, reads a handful of pages, and writes a summary naming two to five businesses. Being named comes down to four things: 1. Being readable. If the page's content only appears after JavaScript runs, many crawlers see an empty shell. Server-rendered or prerendered HTML is a prerequisite, not an optimisation. 2. Being unambiguous. The model needs to know what you do, where you do it and that you are the same business it saw elsewhere. Identical name, address and phone number everywhere on the web is the entity signal that ties those references together. 3. Being quotable. Short, factual, self-contained sentences get lifted verbatim. Long paragraphs of marketing adjectives do not. 4. Being corroborated. A claim on your own site is weak. The same fact on a chamber of commerce listing, a supplier's stockist page, a local news mention or a directory is what turns it into something the model will repeat.

Answer-first structure, in practice

Put a forty to sixty word direct answer immediately under the heading, before any preamble. Write every H2 as a question a customer would actually ask. Keep paragraphs to two to four sentences. Prefer bullets to tables — tables are frequently mangled when a model parses a page. This is not a trick. It is the same structure that helps a customer skim, which is why it works.

The markdown mirror

Alongside your normal pages, publish a clean plain-text or markdown copy of each key page — no navigation, no scripts, no styling — and list them in your /llms.txt. Agents that struggle with a rendered page will read the mirror without difficulty. It costs nothing to generate once your site can build it automatically, and it removes the most common technical reason for being skipped.

Measuring something that produces no click

Traditional analytics will not show you this, because a person who reads your name in an AI answer and rings you never appears as a website visit. Three things worth tracking instead:

• Presence. Once a month, run the ten questions your customers actually ask into ChatGPT, Perplexity, Gemini and Google's AI mode. Record whether you are named. That number, tracked over time, is the metric.
• Crawler visits. Your server logs show GPTBot, ClaudeBot, PerplexityBot and Google-Extended arriving. If they never arrive, nothing else matters yet.
• Unattributed enquiries. A rise in calls from people who "saw you online" but cannot say where is, increasingly, this channel.

Four myths worth clearing up

• "Schema alone gets you cited." It helps machines parse you. It does not make thin content worth quoting.
• "You need to publish constantly." Depth beats frequency. One genuinely complete page on a topic out-performs six shallow ones, and shallow duplicates actively hurt.
• "Blocking AI crawlers protects your content." It removes you from the answers. If you want the recommendations, allow the bots in robots.txt.
• "It is too early for a small Irish trade." The queries where local businesses get named — "best X near Y", "who does Z in Wexford" — are exactly the ones answer engines handle most confidently. Small local sites are competing on structure, not budget, which is unusually favourable ground.

In our experience

The change that has moved citations most reliably for our clients is the least glamorous one: rewriting the top of each service page as a plain, factual answer to the question the page is named after, then making sure the same business details appear identically on the site, the Google profile and every directory listing.

What is GEO in plain English?

Generative Engine Optimisation is the work of getting your business named inside AI-written answers rather than just ranked in a list of links. It overlaps heavily with good SEO — clear structure, accurate facts, technical crawlability — with more emphasis on being quotable and being confirmed by sources other than yourself.

Do I need a separate strategy for AI search and Google?

No, and treating them separately usually wastes money. The same foundations serve both: crawlable HTML, one clear topic per page, question-shaped headings, consistent business details and genuine third-party mentions. The difference is in how you measure success, not in what you build.

How do I know if AI tools can read my website?

Ask for the page source and check whether the main text is present before JavaScript runs; a page that shows an empty container is invisible to many crawlers. Then check your robots file is not blocking GPTBot, ClaudeBot, PerplexityBot or Google-Extended, and look in your server logs for those bots arriving.

What is an llms.txt file and does it matter?

It is a plain-text file at the root of your domain that summarises the business and lists your key pages for AI agents. It takes minutes to write and is a reasonable signal, but it is a directory, not a ranking factor — it helps a crawler find good pages, it cannot make weak pages worth citing.

How long before AI search sends me work?

Slowly and then suddenly, in our experience. The technical groundwork can be done in a fortnight, but citations tend to appear once third-party corroboration builds, which is a matter of months. Treat it as compounding visibility rather than a campaign with an end date.

Who wrote this

Joey Bray, founder and principal engineer at Digital Bridge in Gorey, Co. Wexford. He scopes and builds every project personally and writes these guides from the work itself. Questions on anything above: 085 224 1848 or [email protected].