ChatGPT marketing, visibility and ads
A growing share of your buyers now ask an assistant before they ask a search engine, and the reply names two or three businesses. We work on whether yours is one of them, on the sponsored space underneath, and on what happens to the people who arrive already half sold.
Somebody in your area is describing your job to ChatGPT right now and being handed a shortlist. You do not need a marketing department to be on it. You need the few facts about you that a machine can read, verify and repeat.
Three minutes, twelve questions. No email needed to see or copy your result.
Who should I use for this, and who is actually any good?
Based on what I can find, three would suit you:
Where that answer came from
What we work on around ChatGPT
One job with three parts: be findable, be worth naming, and be able to prove it happened.
Where we come in
Nothing has broken. The traffic charts look much as they did. The gap is somewhere the charts do not go.
You have not lost anything you can point at, and that is exactly what makes this hard to notice.
Wrong service, an old address, a director who left in 2021, or a confident sentence about something you have never sold. Nobody at OpenAI wrote that. It was assembled from what the internet says about you, and most of the internet has not been updated in years.
People arrive knowing your prices, your process and the name of your competitor, and nobody can explain where they got it. Analytics shows direct traffic and a branded search. The conversation that decided it happened somewhere you cannot see.
ChatGPT Ads reached UK advertisers this year and the honest answer for most businesses is not yet, or not until the free work is done. That answer is worth more than a rushed campaign, and it is the one we would give you on a first call.
Three names come back and none of them is yours. Not because the work is worse, but because there is almost nothing written about you anywhere except your own site, and a machine treats a business that only describes itself as a business nobody can vouch for.
A hero line about craft and passion, prices on request, the area you cover implied rather than stated. A person will work it out. A machine assembling an answer will move on to the business that put the facts in a sentence.
Good news: volume is not the lever here. A handful of accurate, specific, verifiable facts in the right five or six places will do more for you than a year of posting, and once they are done they mostly stay done.
First, a definition
They get discussed as one subject and they are not. They need different budgets, different people and different measures of success, and mixing them is why so much of this feels vague.
The unpaid one. Whether an assistant names you when somebody describes their problem, and whether what it says about you is true. Slow, cumulative and the closest thing here to the old rules of search.
ChatGPT Ads. Labelled sponsored placements shown below an answer, sold through OpenAI's own Ads Manager. New, thin on public data, and genuinely useful for a narrow set of businesses today.
Not marketing at all, strictly, but it is usually the thing that pays for the rest. The unglamorous hours: quoting, drafting, summarising, answering the same enquiry for the ninth time this week.
We do all three. Most first conversations end with us recommending the first and the third, and asking you to wait on the second.
The answer lab
Type the question, your name and the sentence you use to describe what you do. The panel shows the shape of the reply, where it drew from, and what the sponsored slot underneath would look like. Then it checks your sentence against five things that decide whether a machine can use it. Nothing leaves this page.
Use the words a customer would use, not the words your industry uses. That difference is most of the job.
How much is written about you elsewhere
Everything you type is processed in your browser. It is not stored and never reaches TWV Media.
Who can help me with this, near me?
Assistants change how they present answers constantly, and two people asking the same question can be answered differently. The shape is the point here, not the pixels.
The AI visibility scorecard
It scores whether a machine can find you, verify you, read you and pass you a customer. Your answers stay in this browser and disappear when you close the tab.
What we actually do
Take the lot or take the part that is currently broken. Either way one senior person stays responsible for the thread, and you get told plainly when something is not worth doing yet.
We ask the assistants what a buyer would ask and record what comes back about you, your competitors and your category. Wrong facts, missing mentions and the sources doing the talking. It is usually an uncomfortable half hour and the most useful document you will read this quarter.
The plain facts in plain text: what you sell, who for, where, what it costs, what happens next. Structured data for the organisation, the services, the prices, the reviews and the questions. Crawler access checked rather than assumed.
Directories that still matter, trade and local press, supplier and partner pages, review platforms, membership and accreditation listings. A machine trusts a claim more when three unrelated sources make it. Nothing invented, nothing bought from a link farm.
Pages that answer the actual question in the buyer's words, in the first paragraph, with the specifics attached. Comparisons, prices, timescales, what you do not do. Written for a person, structured so a machine can lift it whole.
Account setup, the context hints that decide when you appear, creative that reads as helpful rather than shouted, the pixel or the conversions API, and a fair comparison against the same money spent on search or social. Including the answer that it is not for you yet.
Assistant referrals separated in analytics, a source question on the enquiry form, branded search watched as the shadow of everything else, and a monthly note on what the assistants now say about you. Undercounting is inevitable here. Pretending otherwise is not.
The paid side
It is a real advertising platform now, open to UK businesses, and it is also about eighteen months younger than the confidence with which some agencies discuss it. Here is what is actually true, and what we would say to you about it.
Accurate to the best of our knowledge as of September 2026. This platform has changed materially every few weeks since launch, so treat any written summary, including this one, as a starting point rather than a specification. We check the current position before anybody spends.
The interesting question is not whether ChatGPT Ads work. It is whether they beat what the same money already does somewhere else, for you, this quarter.
Where we would say yes today: a clearly defined product with a decent margin, a landing page that already converts, tracking that already works, and a category people describe conversationally rather than search for by name. That is a real set of businesses, and it is a smaller set than the one being pitched.
Where we would say wait: if the assistants are currently describing you wrongly, paying to be shown beside a bad answer is an expensive way to confirm it. Fix the free half first. It is usually a few weeks of work and it does not stop when the budget does.
Where we would say wait: if you are a one-person business, your first pound is almost always better spent making the facts about you findable and correct. Advertising into an answer that cannot verify you is paying to be a stranger.
The three-slot arithmetic
In search you can be tenth and still collect something. In an answer there is no tenth. This works out what that means for you, starting from the size of the field you are competing in.
Straight arithmetic. The baseline assumes three slots shared evenly across the field, so with twelve credible businesses you would be named in roughly a quarter of answers by chance alone. That baseline is the whole point: every hour of this work is an attempt to sit above it. The first two sliders are assumptions rather than measurements, because nobody publishes that data yet, and the annual figure assumes one enquiry in three becomes a customer.
Assistant traffic will always look too small in analytics and too big in the sales meeting. Both are correct, and only one of them pays wages.
A person asks an assistant, gets three names, reads about you, closes the tab, and searches your name two days later on their phone. Last click records a branded search. The thing that actually decided it left no trace at all. This is not a reporting problem you can solve. It is one you plan around, by putting a source question on the enquiry form and by watching branded search as the shadow of everything upstream.
The first two rows will disagree, usually by a wide margin. Saying so in month one stops it becoming an argument in month five.
How it runs
Findable, then correct, then quotable, then measurable. Doing these out of order is the most common way to spend a budget on this and have nothing to show for it.
Two weeks of asking assistants what a buyer would ask, logging what comes back about you and your competitors, and tracing which sources the answers lean on. You get the transcript, not a summary of it.
Crawler access, structured data, the plain-language facts, the pages that answer real questions. This is unglamorous, largely technical and where most of the movement comes from.
Listings, press, partners, reviews, accreditations. Slower and more human than the technical work, and the part that decides whether a machine treats your claims as verified or merely stated.
The same questions asked again, the source question on your enquiry form, branded search, and a decision about whether paid placement now earns a trial. Then we agree what deserves to continue.
For a business or sole trader who will do the work themselves once they can see it clearly.
We do the readable half and the described-elsewhere half, and prove the measurement works before anything is bought.
For teams who want this run alongside the rest of the marketing rather than as a side project.
Before you ask
Nobody can guarantee that, and you should walk away from anybody who says otherwise. What can be done is remove the reasons you are currently left out: facts a machine cannot read, claims nothing outside your own website supports, a crawler quietly blocked, no page that answers the question in the words a buyer uses. Businesses that fix those get named more often. That is a pattern, not a promise.
It overlaps heavily and the overlap is good news, because most of it is work you would want done anyway. The differences are real though. There is no position ten. Your own website is one voice among many rather than the main event. And an answer is assembled and paraphrased rather than linked, so being quotable matters more than being ranked. We usually run the two together rather than selling them as separate products.
That is the most common reason people arrive on this page, and it is fixable more often than it feels. The wrong fact almost always traces back to a source: an old listing, a stale directory entry, a press piece from a previous version of the business, a page on your own site nobody has looked at since 2019. We find the source, correct it where it can be corrected, and publish something clearer and more recent to compete with what is left.
Often better than it works for a large one, because the field is smaller and the facts are simpler. A sole trader with a clear service, a stated area, real reviews with words in them and a handful of accurate listings can be genuinely competitive in an answer against companies spending far more. Scale is not the advantage here. Legibility is.
Sometimes. It suits a defined product with a decent margin, a page that already converts and tracking that already works. It suits almost nobody as a first move. The platform is young, the audience is capped to the free and Go tiers by design, and there are no benchmarks worth trusting. We will tell you which side of that you are on, and we would rather lose the campaign than set one up that was never going to work.
Three things, in this order. The enquiry form asks how they found you and an assistant is one of the options, which is crude and the single most useful change most businesses can make. Analytics separates visits arriving from an assistant, which will always undercount. And branded search gets watched, because a name typed into Google two days later is often the only visible trace of a conversation that has already done the selling.
Yes. Almost everything that makes you legible to one assistant makes you legible to the others, because they are reading the same web. This page is about ChatGPT because it is the one most of your customers are using and the only one currently selling advertising at any scale, but the baseline audit covers the others and the fixes are shared.
No. The answer lab, the scorecard and the calculator all run in your browser. Nothing is sent to us and nothing survives closing the tab. If you copy your score, that copy goes to your own clipboard and nowhere else. Our privacy notice explains how the rest of the site works.
It depends on how much is broken, how big the site is, and whether you want the coverage work running monthly or done once. After a twenty-minute call you get the scope and the price on one page, with no long contract behind it. If the honest answer is that a few hours of your own time would fix most of it, we will tell you that instead.
ChatGPT and OpenAI are trademarks of OpenAI. TWV Media is an independent agency and is not affiliated with, endorsed by, sponsored by or a partner of OpenAI. Platform details on this page describe publicly available information and change frequently.
Also worth a look
Book a call
Bring your business name and the question a customer would ask. We will run it live, read you what comes back, and tell you the first thing we would change, whether or not you ever hire us.
Book a free 20-minute call