We are webs.ie, a small web and AI studio in Dublin. On 29 August 2026 we asked Google's AI the question a Dublin householder types when the immersion heater packs in: plumber dublin. Then we pulled the list of websites its answers actually quote.
Google's AI Overviews is the summary Google now prints above the ordinary blue links. It has been live in Ireland since March 2025, and most people reading this have been shown one today without thinking about it.
Of the ten sites its answers cite most often, six are job boards or salary sites. Indeed, Jooble, IrishJobs, Glassdoor, careersportal.ie, salaryexpert.com. Exactly one working plumbing business is on the list, in seventh place.
Read that as the owner of any small Irish business with a website that works fine. Google's AI is quoting the sites that answer "how do I become a plumber". Barely any of them answer "who do I hire".
We checked two other trades as well. There is a five-minute check at the end of this article that tells you where your own stands, and it costs nothing to run.
The figures are ours, pulled on 29 August 2026 from DataForSEO's index of AI answers, filtered to Google AI Overviews, Ireland, English: 112 mentions for that phrase.
Then we checked two more trades
Solicitors is sharper. For solicitor dublin, AI Overviews carried 424 mentions in Ireland, against a monthly AI search volume of 138,350 on DataForSEO's estimate.
Not one Dublin law firm is in the top twelve sources. The AI reaches for Citizens Information, the Law Society, Reddit, Wikipedia, Indeed, Facebook and the Courts Service first. The first actual firm, Roe Solicitors, comes in at thirteen, and across the whole top twenty only two cited sites are law firms.
Dentists is the fair counterweight. For dentist dublin, six of the top ten cited sites are real practices, 3Dental and Dental Care Ireland among them. That trade is doing something the other two are not. Four slots still go to Indeed, YouTube, Facebook and a dentistry jobs board.
Two honest limits. This is AI Overviews, not ChatGPT: we tried the same read for ChatGPT and could not get it, because that dataset is United States English only, and when we ran dentist dublin it resolved Dublin to Dublin, Virginia. And these are small numbers, 112 mentions for the plumber phrase, in an Irish index far smaller than the UK's. Small samples move around.
Does any of this cost you money yet?
Partly. Here is the careful version.
Google's AI Overviews do cost clicks on question-style searches — Pew Research Center tracked 68,879 US searches in March 2025 and found people clicked through to a website 8% of the time when an AI summary appeared, against 15% when it didn't — but the Irish picture on ordinary commercial searches is genuinely split: Friday's analysis of 175 million Irish searches found average click-through fell from 3.62% in spring 2025 to 2.67% in spring 2026, while ePresence's look at 55 Irish commercial sites over the same three months a year apart found organic clicks up 0.77%, so a number-one ranking is worth less than it used to be rather than worth nothing.
Those are averages across every result, not a measurement of what first place on its own is now worth.
In plain terms, nobody credible says the web has stopped sending you customers. What changed is that a machine now stands between the question and your front door, deciding who gets named.
Why the machine skips you
Three reasons, and none of them are your fault.
It reads a different page than you see. If your prices, your service list or your opening hours only appear after the page finishes loading itself inside someone's browser, several of these readers never see them at all.The precise version, because this is the part people argue about:
In log studies by Vercel and MERJ (December 2024) and a re-test by OpenSeoTest (January 2026), OpenAI's and Anthropic's crawlers and PerplexityBot only read the HTML your server sends and never run the JavaScript that assembles a page inside a visitor's browser — so any text that appears only after those scripts run can be missed — though Apple's Applebot and the crawler behind Google's Gemini and AI Overviews (live in Ireland since March 2025) do run JavaScript, and none of these companies states its rendering behaviour in its own crawler documentation.
We found the worst example of it on our own blog, while writing this. Every article on webs.ie, this one included, had its words fetched by a script after the page arrived, so GPTBot, ClaudeBot and PerplexityBot read the whole journal as three words. We moved the rendering to the server the same afternoon and re-checked all nine articles using the crawlers’ own user agent. You can check us: Ctrl and U on this page, then search for this sentence.
Job boards publish their facts in a form machines can read. Indeed does not out-write a plumber. It labels every listing with the trade, the county, the pay and the date as structured data, hidden tags telling a machine what each thing on the page is. The trade sites we audit usually leave the same facts as ordinary sentences in a paragraph, with nothing to tell a machine which part is the price and which is the county. Your own security may be the one saying no. Cloudflare, which sits in front of a large share of Irish small-business websites, wrote on 1 July 2026 that "On September 15, 2026, we'll be setting new defaults" for the three kinds of bot it recognises. From that date, on pages that display ads, Training and Agent traffic is blocked by default and Search stays allowed. If that is your site, agents stop at the door and nobody in your business ever made that decision. Cloudflare's AI Crawl Control, free on every plan, shows you who called and lets you decide.We caught one of these on ourselves the same morning. Our AI product's site, sitebrain.app, was turning AI crawlers away with a 403. It was fixed that day.
Some of these visitors have no name at all
The advice everywhere is "block the AI bots", or "allow the AI bots". Both are out of date, because a growing number of them arrive without a name.
We read Cloudflare's verified-bot directory on 29 August 2026. Most entries filed under Agent carry no user-agent string at all: OpenAI's ChatGPT agent, Amazon's Bedrock AgentCore, Browserbase, and xAI's agent, which is called GrokAgent and not any of the other names circulating online. They sign each request with a cryptographic key your site can check instead. Others run on a person's own laptop, like Nous Research's Hermes, wearing the identity of whatever scraping tool they use; we searched two bot registries the same day and found no Hermes entry in either.
Two practical consequences. Google-Extended does not control AI Overviews, whatever you have read: it is an opt-out for training and for grounding Gemini, while AI Overviews are fed by Googlebot, the same crawler as ordinary search. And a robots.txt block only stops the bulk crawlers. Perplexity, Meta and Google all state in their own documentation that the fetchers running because a person clicked something may ignore it.
Meanwhile, Shopify quietly did it for a million shops
Shopify's own developer changelog (posted 5 August 2026, effective 21 August) says every standard Shopify storefront — the Liquid themes nearly all stores run — now exposes tools an AI agent can call, including catalogue search, cart edits and checkout, with nothing for the merchant to install; in practice hardly anything can call them yet, because that same note limits agent support to Chromium browsers in an origin trial and Google's Chrome documentation still ships the feature switched off by default.
Almost nothing can use it today, and that is not the point. A shop owner on Shopify woke up on 21 August with a website an agent can operate. A Dublin dentist, solicitor or garage got no such letter.
Check your own site in five minutes
Nothing to install, and you do not need us for it.
- Ask the machine your customer's question. Open ChatGPT or Google's AI Mode and type what a good customer would type. "Emergency dentist Dublin 8." "Employment solicitor Dublin." Are you named? Who is? A directory and two competitors is your result.
- Read your own page the way a machine does. Open your website and press Ctrl and U together (Cmd, Option and U on a Mac). A window of raw code appears. Press Ctrl and F and paste in a sentence from the middle of your page. No match means that sentence never reached the readers that skip JavaScript.
- Check the doorman. Type your own address followed by /robots.txt, so yoursite.ie/robots.txt, into the browser bar. Look for Disallow lines underneath GPTBot, ClaudeBot, PerplexityBot or OAI-SearchBot. That is your website turning those readers away, usually because a plugin or a host put it there years ago.
If all three come back clean you are in better shape than most sites we audit.
What we do about it
We call it Agent Readiness and it is unglamorous work. Your words move into the HTML your server sends instead of being assembled afterwards. Your services, area, hours and prices get labelled as structured data. Which bots get in becomes a decision instead of an accident. Then we re-test and show you the before and after.
You already know whether step two or step three made you wince. Send us the address and we will do the rest: what Google and the AI assistants can see on your site, what is turning them away, and whether they name you for the questions your customers actually ask. In plain English, with the fixes ordered by what they are likely to do to your enquiries. The first check is free, and if there is nothing worth fixing we will say so.
Agent Readiness, done for you