ChatGPT Ads for Tutors: How to Run a Small, Safe First Test

A parent types a question into a chat window: who can help my daughter with GCSE chemistry, online, on Tuesday evenings? A few years ago that question went to a search engine, a friend or a school noticeboard. Today some people put it to an AI assistant, and OpenAI now documents a way for businesses to pay to appear in those conversations. If you are a tutor, a small tutoring centre or the builder of a learning app, it is fair to ask whether that is a place to find students.

This guide is a plan for a first test, written for people who teach rather than people who buy media. We have no ChatGPT ad results of our own to quote. What we have is OpenAI's public developer documentation, which we read on 5 November 2026, and a way of working that suits a small business: prepare the page, set up measurement, start small, look often, and stop if it does not work. Our opinions are said to be opinions, and the one tutor in the examples, Dana, is invented.

Why a tutor might look at ChatGPT ads at all

Choosing a tutor is rarely a one-click decision. A parent wants to know whether someone understands their child's exam board, whether lessons fit around football practice, whether the price is fair. A chat window invites full sentences, so a visitor who arrives from one often knows exactly what they want: a subject, a level, a time of week. That is the best kind of visitor for a small tutor, because a clear request is easy to match with a clear offer.

There is a second reason. Word of mouth is slow, a marketplace takes a cut and controls the page, and a local group reaches one town. A paid channel is another door, and a door you control is useful even if you open it for a few weeks before an exam season.

Now the limits. We have no figures for how many families ask an assistant about tutors, or how many would click an ad. Nobody should give you a number and call it fact before there is public data. Ads in a conversation are new, so there are fewer examples and the rules can change. And because students must trust you before they book, some families will always prefer to ask a friend. Treat everything below as a way to find out cheaply, not a forecast.

What is public, and what we could not find

OpenAI publishes developer documentation for its ads product. From the pages we could open: campaigns can choose which ChatGPT surfaces show the ads, such as mobile apps or the web in a desktop browser, and which countries or regions to deliver in, from the locations currently available. A campaign has an objective: impressions, clicks or conversions. Conversion tracking uses a small piece of code, a pixel, on your own website, and the documentation lists standard events that include scheduling an appointment, creating a lead, registering and starting a trial.

Equally important is what we could not find. The pages on who may advertise, which countries ads run in, and OpenAI's advertising policies were not available to our tools, so we cannot tell you what they say about education services, children or claims. The documentation we read says nothing about users under 18. Before you spend anything, read the current policies and the terms for your market, and ask OpenAI's support if a rule is unclear. It is the most important step in this guide, and the one we cannot do for you.

Is a test worth it for you?

Do one piece of arithmetic that has nothing to do with ChatGPT. What is a new student worth to you? Multiply the lessons an average student takes by your average price. If a student stays for twelve lessons at forty dollars, that is four hundred and eighty dollars of revenue before your time and costs. This invented figure tells you the most you could afford to pay to find one student. It is a ceiling, not a target; you want to pay far less.

The second check is capacity. If your week is full, a new channel only lengthens a waiting list. Ads make sense when you have empty slots, a new subject to launch, or a seasonal peak you can serve, such as exam preparation in spring.

Fix the page before you pay for a click

The most common reason an ad test fails is the page the click lands on. A parent who clicked for "SAT maths tutor, online" and finds a home page saying "Learn with Dana" will leave. They never saw your offer. So build a landing page that continues the sentence the ad started.

A good landing page is short and specific. Near the top it says who the lessons are for and what they cover: "GCSE and A-level chemistry, one to one, online, evenings and weekends". It shows a price or range, so nobody has to ask, and when you teach, in the visitor's time zone if you can. It says what happens next, ideally a button that books a trial lesson. It has one honest line about you: qualifications, experience, what a first lesson looks like. And it works on a phone, because many parents will arrive on one.

If you teach several things, build one simple page for each thing you advertise; each takes an hour. If you use Tutafy, your booking page can include a free trial lesson option and shows your times and prices; with another tool, the same principles apply.

One practical warning. To measure bookings you must place the pixel on the page where a booking is confirmed, which needs a site where you can add code. A marketplace profile usually does not allow it, and some booking tools restrict custom scripts. Check this first. Without measurement you pay for clicks without knowing which became students, and the test loses most of its value. In that case build a simple page of your own, or choose a tool that lets you add a tracking script.

Measure the booking, not the click

A click tells you someone was curious. A booked trial lesson tells you someone wanted what you offer. Conversion tracking connects the second to the ad that brought the person. OpenAI's documentation describes a pixel: a snippet loaded from OpenAI's servers, started with your pixel ID, after which your pages send named events when something important happens.

For a tutor, few events matter. Pick the one that matches the real step on your site: "appointment scheduled" if families book a trial in a calendar, "lead created" if they fill in an enquiry form, "registration completed" or "trial started" if they create an account to begin. Choose one and use it consistently. The documentation also says custom events cannot be an optimisation goal, so use a standard event if you want the system to optimise for bookings.

Consent and privacy

A pixel is a tracking tool, and in many places a visitor should agree before it runs. OpenAI's documentation provides a consent switch: when consent is false the pixel sends no measurement events, and blocked events are not replayed later. So load or enable the pixel only after a visitor accepts, and honour a decline. The documentation also says the pixel can detect customer details on a page and scramble them in the browser before sending, which is another reason to keep it off pages where parents type private details.

Tell people what you do: your privacy page should name the pixel, say what it is for and how to refuse. Do not add it to pages that show a student's name, notes or progress. Keep it to public pages and the booking confirmation.

Choose a goal that matches what you can measure

OpenAI's documentation names three objectives. Impressions aim to be seen. Clicks aim to send visitors to your page. Conversions aim to find people who will complete your chosen event and, according to the documentation, bill for valid clicks. You can set a fixed bid, or use "maximize results", where the system adjusts the bid within a daily budget you must set.

If your pixel works, pick conversions and let automatic bidding work within a daily budget. If it does not, run a tiny clicks test to check that page and ad make sense, then fix measurement before spending more. The documentation says view-through, where someone sees an ad, does not click and returns later, is reported separately with a fixed one-day window, while cost per action stays click-based. So a click-based number may understate what the ads did; do not read too much into one day.

Start small, start early, and leave it alone

A new advertising system needs time to find who responds. If you change the ad, budget and page daily, it never gets a clear signal and you never learn which change mattered. Begin with one goal, one ad, one page and a small daily budget you can lose for two weeks. Start earlier than you think: if you want students for the new term, begin well before it.

Then do very little. In the first days the numbers will look strange because the sample is tiny. Leave the settings alone, look every two to three days, and keep a diary: date, spend, clicks, bookings. It takes two minutes and is the most useful record you will keep.

Notice what Dana did not do. She did not add five ads on day three, double the budget after one good day, or conclude from two clicks that ChatGPT does not work. She changed one thing, on day nine, because she saw a specific problem. That is the habit worth copying.

Read the results with three questions

When you look every two or three days, you do not need to understand every chart. Ask three questions in order, and fix the first problem before looking at the next.

1. Is it running?

Is the ad approved, in review or rejected? Is today inside your dates? Is the daily budget set and enough for a few clicks at your bid? Has the account hit a limit? Many weak tests were never running properly. If the ad was rejected, read the reason and fix the wording or the page.

2. Is it doing what I expected?

Are people clicking? If not, the ad may be unclear or aimed at the wrong people. Are clicks reaching your page? Check your own site statistics, because a slow page that breaks on a phone loses people unseen. Is the pixel sending events? Make another test booking. If clicks come but bookings never register, the problem is often tracking, not the ads.

3. What is worth changing?

Only when the first two are fine should you optimise. Look for what works: a subject, a wording, a surface. If something is clearly better, add budget in small steps and watch the cost of each booking. Change one thing at a time and wait a few days. If nothing works after a fair test, the honest answer may be that this channel is not right for you now, and that is a useful result too.

Writing an ad a tutor can stand behind

A tutor's ad has one job: help the right family recognise it is for them and take one small step. Think in four parts. First, subject and level, because that is what the family asked about. Second, one honest phrase on what is distinctive: "former exam marker", "dyslexia-friendly teaching", "evenings and weekends". Third, the lowest-risk next step, usually a trial lesson. Fourth, a clear call to action. Keep it short and sound like a person, not a brochure.

If you can add an image, use a real one, such as a worksheet you wrote yourself, and avoid stock photos of smiling children. If the system offers to rewrite your text automatically, read the result before accepting: a rewrite that sounds impressive may promise more than you can deliver.

If you test two versions, keep the difference small: the same page and subject with a different opening line. And one ad is fine for a first test. A clean result from one honest ad is worth more than a confusing result from six.

What not to promise, and what to check about children

Tutoring sits where a careless sentence can cause trouble. Three areas need care.

Results. You cannot control a student's effort, an exam board's marking or a bad day. "Guaranteed" grades, "raise your score by two levels" and "pass or your money back" invite disappointment and, in many places, misleading-advertising rules. Describe what you do and what a first session is like. If you quote a past result, make sure it is true, typical enough not to mislead, and used with permission.

Children. Many students are under 18, and parents often search on their behalf. Advertising to children and collecting their information is regulated in many countries. In the United States, for example, COPPA limits collecting personal information from children under 13 without parental consent; the European Union and the United Kingdom have their own rules. This is general information, not legal advice. Aim ads at parents and older students, do not ask for a child's details on the first page, and ask a lawyer or regulator if unsure.

The platform's policies. We could not read OpenAI's advertising policies, so we cannot say what they require for education services or minors. Read them before you submit. If an ad is rejected, do not resubmit the same text: read the reason, change the wording or page, and ask support. Use a family's testimonial only with clear permission, and keep it accurate.

The checklist before you start

After the busy season

Tutoring has seasons: exam preparation in spring, a catch-up push before the school year, the new-year wave. Whatever the test shows, the next step is the same: look at your diary, write down what you learned, and decide on purpose.

If it brought students at a cost you can accept, continue with the same ad and a slightly larger budget, and plan the next test for the next peak. If it brought clicks but no bookings, return to the page and the measurement; most improvement is there. If it brought nothing, stop, note why, and put the money into something that works for you, such as a referral programme or a better page. A test is only wasted when you do not write down what it taught you.

One thought for the long term. Ads rent attention; they do not build it. Families who stay, and bring friends, come from the quality of your lessons and how easy you make it to book and show up. Use ads, if at all, to meet people who would not otherwise find you.

Pages used, read on 5 November 2026; they show no date. The plan, examples and Dana are ours and invented. We could not open the ad policy, who-can-advertise or where-ads-run pages, so none is described. Nothing from OpenAI's advertiser training is used. Sources: OpenAI developers: Measurement Pixel · OpenAI developers: Supported events · OpenAI developers: Bidding and budgets · OpenAI developers: Platform targeting · OpenAI developers: Location targeting.