AI SEO for Plumbers: Get Named by ChatGPT and Google AI Overviews

A plumbing company gets named for emergency searches by getting the basics right: a complete, accurate Business Profile, one clear page per urgent job, recent genuine reviews and listings that agree with each other. Google says AI Overviews need no special optimisation, so test what each assistant says about you, fix the gaps, then test again.

Nobody can promise you a mention. Answers shift with the searcher's position, the wording and the time of day, and neither Google nor OpenAI publishes a formula for local recommendations. What you control is the evidence an assistant can find about your firm. What you can measure is whether you are named, described correctly and linked to your own site. That is why the test comes before any fixing.

Follow me on Instagram@sagnikteaches

I'm assuming a firm with one to ten vans, a website you can edit, a Business Profile you can log in to, and about six hours a week to spare for a month. If an agency looks after your website, hand them your test sheet and the page outlines below.

Connect on LinkedInSagnik Bhattacharya

Run the emergency test first: 12 queries, three surfaces

Fixing before measuring wastes the first month. A firm that blames its review count may find the real cause is an opening-hours clash between profile and website, or a security setting that keeps ChatGPT's crawler out. So write down what the assistants say today.

Subscribe on YouTube@sagnikteaches

Use three surfaces: Google's AI Overviews (shown on some searches, not all), Google's AI Mode, and ChatGPT with search switched on. Run the same twelve queries on each, which gives 36 checks. Where Google shows no AI Overview, record "none shown" instead of rewording the query until one appears.

• Type each query the way a stressed customer would, and put a real town from your service area where you see [town]. Repeat a few with a second town, because distance is one of the three factors Google names for local results.
• Open a fresh ChatGPT chat for every query and keep your own chat history and saved memory out of it, because both can colour an answer.
• Do not test only from your depot. A result seen from your yard tells you about your yard. Repeat three queries from a phone in a different part of your service area.
• Run one session in the evening. Emergency searches happen at night, and an answer that says "closed" at 9pm is the one that costs you a job.
• Screenshot every answer with the date. Answers change, and without a baseline you cannot tell whether anything improved.
• Repeat three of the queries three times each, back to back, to see how far an answer moves on its own.

1. emergency plumber in [town]
2. burst pipe tonight, who can come out in [town]?
3. plumber open now in [town]
4. blocked toilet plumber same day [town]
5. no hot water, plumber today [town]
6. how much does an emergency plumber cost in [town]?
7. best emergency plumber in [town] with good reviews
8. leak under kitchen sink, plumber today [town]
9. 24 hour plumber [town]
10. leaking water heater, who can fix it today in [town]?
11. blocked drain or sewer backup, emergency plumber [town]
12. [your firm name] reviews

Score each check on four things. Was your firm named? Were its details right (phone, hours, service area)? Did the answer link to your own website or only to someone else's page for you? Which competitors were named instead? Four columns are enough:

In ChatGPT, three follow-ups are worth asking after the first answer: "Which sources did you use for this?", "Which of these is open right now, and how do you know?" and "What would make you confident recommending [your firm name]?". Treat the replies as clues. An assistant's account of its own reasoning is not proof of how it actually chose, but the sources it lists are real pages you can go and read.

A worked baseline for a four-van firm: 36 checks, 10 mentions

Take a four-van plumbing firm with 61 Google reviews. Every figure in this example is invented. Its nearest competitor, "Competitor A", has about 380. The owner ran the twelve queries above on three surfaces in one evening session.

Ten mentions looks respectable until you split it. Three came from the firm's own name query (query 12), so only seven of the 28 non-brand answers named the firm. The pattern behind them is what matters:

• Open now: 0 of 3. The Business Profile said 8am to 6pm, the website said "24-hour call-outs", and one directory said 7am to 5pm. No surface named the firm for this query. On the "24 hour plumber" query, ChatGPT did list it, but last and with the note "hours conflict, call to confirm", behind two firms with unambiguous hours.
• Water heater, blocked toilet and blocked drain: 0 of 12 checks. The site had one "emergency plumbing" page and a bullet list of other services. Nothing existed for an assistant to quote on those jobs.
• Cost question: 0 of 3. The answers cited competitor pages that stated a call-out fee and how labour is charged. The firm's site said "competitive prices".
• Five wrong details. Three wrong hours, one old phone number copied from a directory entry nobody had updated, and one answer saying the firm did not cover a town it has served for years.
• Zero of ChatGPT's five mentions linked to the firm's own website. All five cited directory pages. That clue led the owner to the website's security plugin, where a switch labelled "block AI crawlers" was on and had turned away OAI-SearchBot.

The owner ranked the fixes by effort, not by importance. Switching off the crawler block took ten minutes. Making hours, phone and service area agree on the profile, the website and three directories took about ninety minutes. Two new pages and a cost section took about eight hours across a week. The review habit costs fifteen minutes a day and never ends.

At day 30, using the same queries at the same time of day, the firm was named in 17 of 36 checks, had one wrong detail, and ChatGPT linked the firm's own site in six of its eight mentions. Competitor A was still named in 20 of 36. The blocked-drain queries were still at zero, because Search Console's URL inspection showed the new page as indexed only on day 26. First-place mentions went from two to four.

That is a good month, and still not proof of anything permanent. Seven extra mentions across 36 checks sits close to the run-to-run noise you measured, so the owner repeated the sheet at day 60 before deciding the plan worked. The more reliable gains were the accuracy and the own-site links, because those have specific causes you can point to.

What Google and OpenAI actually say about getting included

Strip away the advice from agencies and a few primary statements remain. Google's page on AI features says a page must be indexed and eligible to be shown in Search with a snippet, and that there are no additional requirements to appear in AI Overviews or AI Mode, no special optimisations and no special schema.org structured data. It also says indexing and serving are not guaranteed. The usual fundamentals still apply: let crawlers in, link pages internally, keep important content as text and make any structured data match what is visible. Google's AI features page is short enough to read in five minutes.

For local results, Google's help page names three main factors: relevance, distance and prominence. Prominence means how well known a business is, and the page ties it to how many websites link to you and how many reviews you have, adding that more reviews and positive ratings can help. It also says there is no way to request or pay for a better local ranking. The local ranking page is the plain-English source for the three factors.

OpenAI's crawler documentation separates three agents. OAI-SearchBot is the one used to surface websites in ChatGPT's search features, and OpenAI recommends allowing it in robots.txt. GPTBot relates to training generative AI models. ChatGPT-User acts on a person's request and is not used to decide whether content may appear in search. Changes to robots.txt can take about 24 hours to take effect. OpenAI's bots page lists all three.

What none of the pages publish is a ranking formula for a local recommendation inside ChatGPT. Third-party analyses disagree about which directories and maps ChatGPT leans on, and those analyses date quickly. So the sensible reading is: an assistant can only describe you from what it can read, so make sure it can reach your pages, find the same facts in several places, and find proof that you do the job well. Everything below follows from that.

Business Profile fixes that matter at 11pm on a Sunday

Your Business Profile is the one asset every local searcher and most assistants can see, so it has to be boring and exact. Google's guidelines are specific about several fields that plumbers routinely get wrong.

The hours row deserves its own warning. If a human answers your phone only until 6pm and an answering service or voicemail takes over after that, say so in the same words on the profile, the website and the directories. "24-hour emergency line" and "open until 6pm" can both be true, but an assistant sees two conflicting facts and hedges, which is what happened to the four-van firm above. If you are not genuinely able to attend at night, do not claim it: a customer who rings at 11pm and gets voicemail will leave a review that every assistant then reads. My tutorial on optimising a Business Profile with AI covers the remaining fields.

One practical check: search your own firm name and read the profile as a stranger would, on a phone. If the first thing you see is a gap, an old holiday notice or a phone number you no longer answer, an assistant is reading the same thing.

The burst-pipe page: what a stressed homeowner needs in 150 words

Most plumbing sites have one emergency page written to impress, not to answer. Here is a typical opening and a rewrite. The numbers in the rewrite are placeholders: use only figures from your own job records.

Before: Welcome to our family-run plumbing business. We offer a wide range of plumbing services at competitive prices. Contact us today for all your plumbing needs.

After: Water coming through a ceiling or from a split pipe? Close the main stop valve first, then call [phone]. We answer 24 hours a day, every day, and send the nearest of our four vans to [Town A], [Town B] and [Town C]. The call-out fee is $[x], which includes the first half hour; after that labour is $[y] per half hour plus parts, and we give you the price before any repair starts. Text photos of the leak to [mobile] and we will tell you what to expect before we arrive.

The rewrite works because it answers the questions an assistant is trying to answer on the customer's behalf: what is this, who does it, where, when, for how much, and how do I reach them. Build the whole page in this order:

1. The first 100 words: what to do right now, your phone number as text (not inside an image), hours, towns covered.
2. A response promise you can keep: pull the median arrival time from your last 50 emergency jobs and state that, not an aspiration.
3. How charges work: call-out fee, hourly or half-hourly rate, what is included, and when you quote a fixed price. Competitors who publish this are the ones assistants cite on the cost query.
4. The jobs you attend: burst pipes, leaking joints, blocked toilets, no hot water, each as a short paragraph with what you do and when it needs a second visit.
5. What you need from the customer: where the stop valve is, a photo, access, a safe place for your van.
6. Proof: licence or registration numbers where your trade makes them public, insurance in a sentence, three real job write-ups with a date, a cause and a fix, and genuine photos.
7. Contact: a tap-to-call button, a text number and a short form, all showing the same phone number.

Use AI as an editor here, not a ghostwriter. Paste your page text into ChatGPT or Claude and ask it to read as a homeowner with a leak:

You are a homeowner with water coming through the kitchen ceiling at 10pm. Read the page text below and list, in order, every question you would still need answered before you rang this firm. Then list anything on the page that is vague, conflicts with itself, or could not be checked. Do not rewrite the page.

[paste page text]

A typical reply (illustrative) lists missing items such as "no mention of the stop valve", "no price or call-out fee", "hours appear in the header image only" and "phone number not repeated at the bottom". Those are real gaps. Be careful with anything the reply says about your business that you did not supply: if it suggests a guarantee period or a response time, that is invention, and it stays out unless it is true.

The same approach works for the pages assistants could not quote in the baseline. Write one short page for each of: water heater faults, blocked toilets, and blocked drains. Each needs its own opening, call-out information, what you do and a real example. Do not clone one page across ten towns with the name swapped. The tutorial on service-area pages without duplicate content shows how to handle towns honestly.

Reviews an assistant can quote, and replies that add facts

Review count and ratings feed the "prominence" side of local ranking, and review text is the most detailed public description of what you are like to deal with. Plan on an assistant reading it.

The rules are strict and specific. Google's review rules prohibit offering incentives such as discounts for reviews, selectively soliciting positive reviews, pressuring customers or asking for specific content. Google's own guidance says you may ask customers to visit a review link or scan a QR code. So the compliant process is to ask every emergency customer, without screening, once the water is back on and the van is leaving:

Hi [first name], it's [your name] from [firm name]. Glad we could get your water sorted today. If you were happy with the work, a short Google review would help other people choosing a plumber: [review link]. Thanks again.

Notice what is missing: no request to mention the service, the speed or a technician's name, and no reward. Reviews that mention a burst pipe at 10pm tend to arrive anyway when you ask within a day, while the job is fresh. My tutorial on review follow-ups for trades covers timing and automation.

Replies are the other half. Google's guidance says replies are public, should be short and professional, and should not be the same generic thank-you for everybody. It also says not to use replies to advertise deals and not to share a reviewer's private details. A reply that adds a fact helps both the reader and an assistant:

Thanks, [first name]. Glad we could reach you quickly and get the burst pipe under the stairs fixed in one visit. If the new valve gives any trouble, ring the office and quote the job reference on your invoice. [Your name], [firm name]

Draft replies with AI if it saves time, but you edit every one and you post it yourself. Nothing should auto-post: the tutorial on responding to Google reviews with AI explains why.

You can also mine your own reviews for the words customers use. Export or copy your last 40 reviews into ChatGPT or Claude, with customer names removed, and ask:

Below are 40 customer reviews of a plumbing firm. List the five most repeated things customers praise and the three most repeated complaints. For each, give the number of reviews that mention it and quote one short phrase. Say clearly if you are unsure of a count.

[paste reviews]

An illustrative result: "arrived quickly" in 9 reviews, "explained the cost before starting" in 7, "tidied up" in 5, and, on the negative side, "price changed from the phone quote" in 3. Do not trust the counts without checking: language models miscount, so search the pasted text for each phrase yourself. The useful output is the pattern. If the price-change complaint is real, fix it on the page by stating how phone quotes work. If "arrived quickly" is real, your response-time promise has evidence behind it, and you can describe it truthfully.

Listings that let an assistant cross-check your details

An assistant that finds the same name, phone number, hours and service area in several independent places has less reason to hedge. The jargon for this is NAP consistency (name, address, phone), and it extends to hours and service area. Here is what I would claim and check, in rough order of effort per benefit:

Keep one master record, a simple note holding your exact name, phone, hours, service area and website, and copy from it every time. When you change anything, such as a new mobile number or holiday hours, change all listings in the same sitting. A mismatch audit takes about an hour: search your phone number in quotes, then your firm name plus "plumber", and write down every page that shows a different number or hours.

FAQs, robots.txt and markup: what to write, what to skip

Write FAQs for the customer. Google ended FAQ rich results in May 2026, so an FAQ block will not earn extra decoration in results, and Google says AI Overviews need no special markup. A good FAQ still gives assistants a clean, quotable answer. Pick six to eight real questions from your phone log: "Do I need to switch the water off?", "Is there a call-out fee?", "Will you fix it in one visit?", "Can I send photos?", "Do you work on Sundays?" Answer each in two or three sentences with your actual policy. If you cannot state a policy, that is a business decision to make first, not a content problem.

Check robots.txt in two minutes. Open yoursite.com/robots.txt and read it. A file like this lets all crawlers in and points to your sitemap:

User-agent: *
Disallow:

Sitemap: https://www.example.com/sitemap.xml

If you see lines that disallow OAI-SearchBot, or a blanket disallow, ask whoever manages the site why, and look at any security or SEO plugin with a setting that blocks "AI crawlers". Blocking GPTBot (training) is a separate choice from blocking OAI-SearchBot (search). If you change anything, allow roughly a day before re-testing, which matches OpenAI's own guidance.

Structured data is optional. If your web person wants to add it, schema.org has a Plumber type, a child of HomeAndConstructionBusiness and LocalBusiness, which supports name, telephone, areaServed, opening hours and more. Place a block like this in the page head, inside a script tag of type application/ld+json, and make sure every value matches what is visible on the page:

{
  "@context": "https://schema.org",
  "@type": "Plumber",
  "name": "[firm name]",
  "url": "https://www.example.com/",
  "telephone": "[phone]",
  "areaServed": ["[Town A]", "[Town B]", "[Town C]"],
  "openingHoursSpecification": {
    "@type": "OpeningHoursSpecification",
    "dayOfWeek": ["Monday","Tuesday","Wednesday","Thursday","Friday","Saturday","Sunday"],
    "opens": "00:00",
    "closes": "23:59"
  }
}

Only use the 24-hour block if a human or a properly configured service answers every hour. The markup helps nobody if it contradicts the profile.

Thirty days from baseline to review habit, week by week

The hours below are estimates for a firm without an in-house marketer, so adjust them. The sequence matters more than the hours: measure, fix what is cheap and wrong, build the missing pages, then build the review habit.

If you only have two hours, do the crawler check, the hours fix and the phone-number audit. They cost almost nothing and they remove the most embarrassing errors.

Re-testing: how to read the second round fairly

Run exactly the same queries, on the same surfaces, at the same time of day, from the same places. Then compare four numbers: mentions, wrong details, own-site links and first-place mentions. A gain that appears on a different query from the one you worked on is probably noise; a gain on the query you built a page for, with that page linked in the answer, is evidence.

Add independent measurements so you are not relying on screenshots alone. Google announced Search Generative AI performance reports for Search Console in June 2026, and every site had them by 31 August. They show impressions only, broken down by page, country, device and date, with AI Overviews and AI Mode counted together: no clicks and no queries. The ordinary Performance report still shows clicks, and Google's AI features page says traffic from AI features is counted there. Bing Webmaster Tools added an AI Performance report in public preview on 10 February 2026: it shows how often your pages are cited as sources in Microsoft Copilot, Bing's AI summaries and some partner services, plus a sample of the key phrases the AI used when retrieving them. Neither replaces the test sheet, because neither tells you whether the answer was correct.

The cheapest measure is the most reliable: ask every new caller how they found you, and add "an AI assistant" to the answer list on your booking form. Pair it with a look at your website analytics for referrals from assistants, remembering that many people read your name in an answer and then ring you directly, so any referral figure undercounts. The tutorial on tracking whether AI search sends customers goes further.

Decide in advance what would change your mind. If after 60 days the pages are indexed, the details are accurate and the own-site links are present but you are still unnamed on broad queries such as "emergency plumber in [town]", the likely cause is prominence: reviews and links accumulate slowly, and a competitor with 380 reviews has a head start no page can close in a month. That is a reason to keep the review habit going, not to abandon the work.

Shortcuts that backfire on emergency searches

• Mass-producing town pages with AI. Google's spam policies, updated on 28 August 2026, list generating many pages with generative AI without adding value as scaled content abuse. Ten near-identical "plumber in [town]" pages fit that description.
• Buying or inventing reviews. Google prohibits fake engagement and reviews posted at the request of one person from multiple accounts. A removed batch of reviews is worse than a low count.
• Paying for a guaranteed ChatGPT ranking. Google's own guidance on hiring an SEO says no one can guarantee a number one ranking, and to find someone else if they promise first place. The logic applies to AI answers too.
• The four covered above. A keyword-stuffed profile name such as "Best 24hr Emergency Plumber [Town]", a 24-hour claim nobody staffs, a one-click "block AI bots" setting and automated posts or replies. Each one is quick to commit and slow to undo.

If you also buy leads through Google, the move of Local Services Ads towards Performance Max is a separate decision from everything above, and it can wait until the free fixes are done. For the bigger picture of how AI answers are changing what local searchers see, read how AI Overviews are changing local search for small businesses.

Questions plumbers ask about AI search

Do I need an llms.txt file or special markup to be named?

Google says AI Overviews and AI Mode need no new machine-readable files, AI text files or special schema.org markup. OpenAI's crawler page asks site owners to allow OAI-SearchBot in robots.txt and does not ask for an llms.txt file. Spend the time on accurate pages, consistent listings and reviews instead.

Should I block GPTBot to protect my site?

GPTBot and OAI-SearchBot are separate. OpenAI describes GPTBot as the crawler behind model training and OAI-SearchBot as the one that surfaces sites in ChatGPT search. Blocking GPTBot does not remove you from search answers, but blocking OAI-SearchBot can. Decide each on its own, and allow about a day for a robots.txt change to take effect.

Can I pay to be recommended in an AI answer?

Google's help page says there is no way to request or pay for a better local ranking, and its guidance on hiring an SEO says no one can guarantee a number one result. Paid placements such as Local Services Ads or search ads are separate paid products. Be wary of anyone selling a guaranteed spot in ChatGPT or AI Overviews.

How soon will changes show up in the answers?

There is no fixed timetable. OpenAI says a robots.txt change can take about 24 hours to apply, but new pages must be crawled and indexed, and reviews build slowly. Re-test at 30 days, compare with your baseline, and treat one good week as noise until the same pattern appears at day 60.

Is it worth hiring an agency for this?

It can be, if they explain every change, show how they will measure it and do not promise rankings. Google's own advice is to ask what results to expect, over what timeframe, and whether the advice matches its official guidance. The test sheet in this tutorial gives you a baseline to hold any agency to.

Further reads

• How to Get Your Business Recommended by ChatGPT and AI Search — The general version of this for any local business
• AI Receptionist for Plumbers: Cost, Burst-Pipe Rules and Booking Limits — Once AI sends the call, someone has to answer it
• Can AI Triage Emergency Plumbing Calls Out of Hours? — How to handle out-of-hours emergency calls safely
• Plumbing Lead Generation: Cost per Job Tracked With AI Summaries — Compare what each lead source costs per booked job
• Best AI Overviews Tracking Software: What a Local Business Needs — Paid tools that track AI mentions, and when they pay off
• How to Choose an SEO Agency That Understands AI Search — Questions to ask before paying an agency
• What Are the Best AI Visibility Products for a Local Business? — Which AI visibility tools suit a single-site local business, what each one tracks, and a free manual check to run first.
• AI SEO for Restaurants: Can You Get Recommended by ChatGPT? — How a restaurant can test what ChatGPT and Google AI say about it, then fix listings, menu text, reviews and crawler settings, and what stays out of its hands.
• AI SEO for Roofers: Hire an Agency or Do It Yourself? — Agency or DIY for a roofer's search and AI visibility: pitch claims to check, break-even sums, a weekly rota, a ChatGPT test sheet and contract clauses.
• AI Marketing for Plumbers: A Worked 12-Month Posts and Offers Calendar — A worked 12-month calendar for a five-van plumbing firm: frozen pipes, boiler checks, heater age and drain season, with AI drafts and a review step.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Search Central (AI features and your website, spam policies, hiring an SEO, and the June 2026 post on Search Generative AI performance reports), Google Business Profile Help (guidelines for representing your business, local ranking, posts, reviews, review replies and prohibited content), Bing Webmaster Tools announcement of AI Performance (10 February 2026), Apple Business, OpenAI crawler documentation (bots), schema.org (Plumber type).