ChatGPT ads and North East B2B: the maths nobody is showing you

In the last few weeks ChatGPT ads have grown up. Conversion-optimised bidding, average daily budgets, custom audience lists, negative keywords, geographic exclusions, a bulk API. The platform that launched in February as a national brand awareness surface with a self-serve interface now looks a great deal like a performance channel.

Every agency in the country has written that up. What none of them have done is check whether the new geographic controls actually apply to a business in Newcastle, Gateshead, Sunderland or Teesside.

We checked. They do not. Not yet.

That single fact changes the entire calculation for a regional B2B business, and it is verifiable in about two minutes from OpenAI’s own published documentation. This post shows you where to look, prices what the gap costs you, and then shows you the workaround that the new conversion tools make possible. It is long, it has a break-even table in it, and the conclusion is not “book a call to find out more”.

What actually changed, and when

A factual baseline, because a lot of what is circulating is a month out of date.

OpenAI switched advertising on inside ChatGPT on 9 February 2026, initially for United States users on the Free and Go tiers through managed partners. Access widened quickly. Geographic targeting expanded from country level to US state, DMA and postal code on 22 May. UK testing began on 5 June with agency partners, and self-serve access opened across June and July.

Since then the release cadence has been close to weekly. The current feature set includes:

• Conversion-optimised cost per click, known as oCPC, which optimises delivery towards users likely to complete a chosen conversion action while still charging per click
• Conversion measurement through a JavaScript pixel, a Conversions API or both, plus automatic advanced matching
• Geographic targeting and geographic exclusions, where an excluded location overrides an included one
• Custom audience lists, uploaded at 25,000 records or more, usable for inclusion or suppression, with bid multipliers set at ad group level
• Negative keywords
• Average daily budgets alongside lifetime budgets
• A bulk API in limited preview, processing up to 1,000 operations per submission
• AppsFlyer and Adjust integrations for app install measurement
• A product card format carrying price and star ratings

The underlying model has not changed. Ads are labelled sponsored cards placed below the answer, one at a time, and OpenAI maintains that ads do not influence the answer itself. Targeting is still driven by context hints, which are free-text descriptions of the conversations where your product belongs, rather than keyword bids. Ads still serve only to logged-in adults on the Free and Go tiers, and are suppressed in temporary chats, image generation, and sessions flagged as sensitive including health, mental health and politics.

The two-minute check every North East advertiser should run

OpenAI’s advertiser documentation states plainly that campaigns can be targeted by country, region or DMA. Reading that, you would reasonably assume a UK advertiser can now select something smaller than the country.

OpenAI also publishes the full catalogue of targetable locations as a downloadable CSV, linked from its campaign targeting documentation. Every location the platform will accept is in that file, with an ID, a name, a country code and a target type.

Open it and filter the country code column to GB.

You get one row. United Kingdom, location ID 1000076, target type Country.

Now filter the target type column instead. Every entry of type Region belongs to Australia, Canada, New Zealand or the United States. Every entry of type TV Region is a United States market. Every entry of type Postal Code is a United States ZIP. There is no North East England. No Tyne and Wear. No Newcastle. No NE postcode district.

So the honest position in August 2026 is this. Sub-country geographic targeting is real, it works well, and it is available in four countries, none of which is the United Kingdom. A UK advertiser can target the United Kingdom and exclude other countries, and that is the whole of it.

We would encourage you to run that check yourself rather than take our word for it, and to re-run it in a month. OpenAI is shipping weekly and the catalogue is a live file. The day GB regions appear in it is the day the rest of this post needs revising, and it is the single most useful thing a North East business can monitor about this platform.

What the gap costs, in pounds

The North East is home to roughly 2.7 million people out of a UK population of around 69 million, which is about 3.9 per cent. On a CPM buy, that percentage is your yield.

Take a mid-market CPM of £35 per thousand impressions.

• 1,000 UK impressions cost you £35
• Around 39 of them are served to someone physically in the North East
• Your effective cost per thousand North East impressions is therefore about £900

A geography multiplier of roughly 26 times, before you have asked whether that person owns a business, or buys what you sell, or was paying attention.

Narrow it once more. If perhaps one in ten of those regional impressions reaches somebody with buying authority in a business of any kind, you are at around four commercially relevant impressions per thousand paid for. Roughly £8.75 each, on a channel that gives you no query-level data with which to verify any of it.

That is what the missing rows in the location catalogue cost. It is also why the rest of this post is about not buying on CPM.

The tier problem is a SaaS problem, not a North East problem

The standard objection to ChatGPT ads for B2B goes like this. Ads reach only Free and Go users. Senior decision makers are disproportionately on Plus, Pro, or a company-provisioned Business or Enterprise seat. So you are paying B2B rates to reach an audience that has already paid to opt out of seeing you.

That objection is correct, and close to fatal, if your buyer is a demand generation director at a software company. OpenAI reported over 9 million paying business users and more than 7 million workplace seats by early 2026, with 92 per cent of the Fortune 500 as customers. That is precisely the population most B2B SaaS advertisers chase, and it is invisible.

Now describe the North East B2B economy.

Department for Business and Trade figures put the region at 166,510 private sector businesses, the fewest of any English region and about 2.9 per cent of the UK total. Overwhelmingly small and micro. Weighted towards construction, engineering, manufacturing, logistics, trades, professional services, care, and increasingly renewables and offshore.

Ask honestly how many 12-person fabrication businesses in Cramlington hold ChatGPT Enterprise seats. How many two-partner surveying practices in Durham pay for Plus. How many regional contractors have a procurement lead on Pro.

Very few. Which means the filter that destroys the case for B2B SaaS advertisers barely touches the case for most North East B2B advertisers. Your buyers are on the free tier, because AI adoption in the regional SME base is early and budgets are tight. The under-digitisation that has been a strategic disadvantage for this region for twenty years is, on this one narrow question, an advantage.

Two forces pulling in opposite directions. Where you land depends entirely on how you buy, and the tools released in the last few weeks have changed that answer.

The workaround: let oCPC learn the geography OpenAI will not sell you

This is the most useful thing in this post, and it only became possible with the conversion-optimised bidding released this summer.

On a CPM buy you fund every wasted impression. The 961 served outside the North East cost you real money and the £900 effective regional CPM is what you actually pay.

On a CPC buy those 961 impressions cost you nothing. The auction absorbs the geographic waste for free and your exposure narrows to one question: how many of the people who do click are in the wrong place?

On an oCPC buy, something more interesting happens. You nominate a conversion action and the platform optimises delivery towards users likely to complete it. It has no idea where your service area is, because you cannot tell it. But it does not need to know, because out-of-area prospects do not become qualified enquiries. If your conversion signal is clean, the optimisation learns your geography indirectly, through the conversion data, and starts avoiding the places that never convert.

You get de facto regional targeting that the platform does not officially offer, assembled out of a feature built for something else entirely.

The catch is the phrase “if your conversion signal is clean”. Most businesses fire the conversion event on every form submission. Do that here and you teach the system that an enquiry from Reading is exactly as valuable as one from Wallsend, which is the opposite of what you want.

So:

1. Do not fire the conversion event on form submission. Fire it server-side through the Conversions API once the enquiry has been qualified as in-area and genuine.
2. If that is beyond your setup, use a hidden postcode or region field and fire the pixel conditionally on the thank-you page for in-area submissions only.
3. Accept that this needs volume to work. Optimisation algorithms need conversion events to learn from, and a campaign generating three enquiries a month will never exit learning. Below roughly 15 to 20 qualified conversions a month, run plain CPC and rely on copy.

That last point matters more than the first two for most readers, and it is the reason this is a technique for businesses with real budget rather than a universal recommendation. The measurement architecture is doing the heavy lifting here, which is exactly the discipline we bring to the paid work in our growth marketing campaigns service.

Your ad copy is the second geo layer

Whether or not you can run oCPC, the copy has a job the platform will not do.

A headline reading “Structural steel fabrication, Tyneside” does two things at once. It attracts the buyer in Wallsend, and it actively repels the buyer in Reading who would otherwise have cost you three pounds to learn was irrelevant. On CPM that qualifying language saves you nothing, because you have already paid for the impression. On CPC or oCPC every click it suppresses is money you keep.

This is the opposite of standard paid search practice, where you strip friction out of the ad to lift click-through rate and let platform targeting handle qualification. ChatGPT has no platform targeting worth the name for a UK regional advertiser, so the copy carries it.

Three rules follow. Buy CPC or oCPC, never CPM, unless you are genuinely national. Put the geography in the headline rather than the description, because the card is small and the description gets skimmed. And judge the campaign on cost per qualified enquiry, never on click-through rate, because a low CTR with tight regional qualification beats a high CTR that fills your inbox with enquiries from Kent.

Negative keywords, added recently, give you a third small lever. They will not exclude places, but they will keep you out of conversations about the parts of your category you do not serve.

Custom audiences, and why most regional firms cannot use them

The custom audience feature looks, at first glance, like the answer to everything above. Upload a list, target it directly, forget geography entirely.

The minimum list size is 25,000 records.

For a national retailer that is a Tuesday. For a 30-person engineering firm in North Tyneside with 1,200 contacts in a CRM, it is unreachable, and no amount of list hygiene will close that gap. The feature is real, it is genuinely powerful for account-based work, and it is not built for the businesses this post is written for.

Two caveats worth knowing if you are near the threshold. Suppression lists work at the same minimum, so excluding existing customers is subject to the same constraint. And bid multipliers are set at ad group level, which means audience strategy has to be reflected in how you structure the account rather than bolted on afterwards.

Writing context hints for a regional B2B buyer

Context hints replace keywords. You write short descriptions of the conversations where your business genuinely belongs, and OpenAI matches them server-side. You never see the query.

Most people write them like keyword lists, which fails, because the system matches conversational context rather than strings.

Weak: “commercial roofing contractors”

Better: “someone comparing quotes for a flat roof replacement on a commercial unit and trying to work out what a fair price looks like”

Better still, build hints around decision stage and let the copy handle location. Three tiers work well.

• Problem-aware, where the buyer has a problem but has not named a solution. “A facilities manager working out why a warehouse roof keeps leaking after repairs.”
• Solution-aware, where they are comparing approaches. “Someone weighing single-ply membrane against liquid-applied roofing for a commercial building.”
• Vendor-aware, where they are choosing a supplier. “Someone asking what to check before appointing a commercial roofing contractor.”

Run these as separate ad groups. You still cannot see which hint fired, but you can see performance by ad group, so structuring by decision stage remains the only way to learn anything from the reporting you are given.

How ChatGPT ads compare to the platforms you already use

The first column is the one that should stop you. Every established platform lets you draw a circle round Newcastle. ChatGPT, for now, does not, and that difference is worth more than any CPC comparison in the table.

It is also why Microsoft Ads deserves a look before you put money into ChatGPT. Same geographic precision as Google, materially cheaper clicks, a user base skewing towards desktop and business. If your reasoning for testing ChatGPT is that you want to reach people who are not on Google, Microsoft will get you further for less.

Does it break even?

Two scenarios, because the answer now depends on whether you can run oCPC properly.

Plain CPC at £3 a click, with hard qualification in the copy putting 40 per cent of clicks in-region and relevant, and a landing page converting at 4 per cent, gives a cost per enquiry of £187.50.

Well-instrumented oCPC, where clean conversion signals lift the qualified share towards 60 per cent and the optimisation improves landing page conversion to 5 per cent, gives a cost per enquiry of £100.

The threshold has moved. Before the conversion tools arrived, ChatGPT ads did not pay for a regional B2B business below about £15,000 average contract value. With properly instrumented oCPC and enough volume to train it, that falls to roughly £10,000, and £5,000 becomes arguable rather than hopeless.

Every assumption there is contestable and you should contest it with your own numbers. That is the point of publishing the model rather than a verdict.

Note also what the model rests on. Everything hangs on the landing page conversion rate, and 4 to 5 per cent from a cold AI-sourced click is not something a generic services page delivers. If you test this channel, the landing page is the majority of the work, not the campaign, because the traffic arrives with no keyword context and no memory of you. Our piece on why your website is your best, or worst, salesperson is worth reading before you spend anything, and if the page needs building properly that sits within websites and digital platforms.

The bigger point: paid AI visibility is the smaller half

There is a two-tier discovery environment inside ChatGPT now, and the tiers do not favour advertisers.

Free and Go users see organic answers plus a sponsored card. Plus, Pro, Business, Enterprise and Education users see organic answers only. Perplexity has committed to remaining ad-free. Claude carries no advertising. So across the AI surfaces where the most engaged and best-resourced buyers spend their time, the only route to visibility is being cited in the answer itself.

Which makes “should we run ChatGPT ads” downstream of a much larger question. Does ChatGPT mention you at all when someone in your sector asks it who to use?

If the answer is no, buying a sponsored card beneath an answer recommending four of your competitors is a strange way to spend money. If the answer is yes, the ad reinforces a recommendation you have already earned and works considerably harder.

Get cited first. Advertise second. We covered the foundations in is AI damaging your search engine rankings?, and the practical work of getting a business surfaced in AI answers sits under AI, automation and insight.

Ad spend stops working the day you stop paying. Citation authority compounds.

The decision rule

For a North East B2B business, in August 2026.

Test now if all four are true.

1. Your average first contract is worth £10,000 or more.
2. Your buyers are owner-managers or operational leads at SMEs, not enterprise procurement teams on company AI seats.
3. You can instrument conversions properly, ideally server-side, and generate at least 15 qualified enquiries a month for the optimisation to learn from.
4. You have £500 to £2,000 you are prepared to treat as tuition rather than performance spend.

Wait if any of these are true.

1. Your average contract is under £5,000.
2. Your buyers work at organisations large enough to hold ChatGPT Business or Enterprise seats.
3. You cannot yet fire a clean, qualified conversion event, which means oCPC will optimise towards the wrong thing.
4. You are not appearing in organic ChatGPT answers for your core service.
5. Your Google Ads account still has obvious headroom, which for most regional B2B businesses it does.

For a good number of North East businesses the honest answer is still wait, and spend the same money on being findable in AI answers instead. But the gap has narrowed considerably since June, and it will narrow again.

The location catalogue is the thing to watch. OpenAI cannot sell to local advertisers without sub-country targeting, and local advertising is where the volume is, so GB regions will appear in that file. When they do, the entire calculation above changes overnight, and the businesses that spent the intervening months building organic AI visibility and clean conversion tracking will be the ones positioned to exploit it on day one.

Frequently asked questions

Can I target Newcastle or the North East on ChatGPT ads?

Not as of August 2026. OpenAI added region, DMA and postal code targeting during the summer, but its published location catalogue contains sub-country entries only for the United States, Canada, Australia and New Zealand. The only GB entry is United Kingdom itself. You can verify this by downloading the location CSV linked from OpenAI’s campaign targeting documentation and filtering by country code.

What are geographic exclusions worth to a UK advertiser?

Less than the coverage suggests. Exclusions operate on the same catalogue as inclusions, so a UK advertiser can exclude other countries but cannot exclude parts of the UK. The feature is genuinely useful in the US and largely academic here until GB regions are added.

How much do ChatGPT ads cost in the UK?

CPC campaigns typically clear between roughly £2.30 and £3.80 per click. CPM campaigns run from about £19 to £45 per thousand impressions, having softened considerably from the fixed launch rate as inventory has scaled. There is no minimum spend.

What is oCPC and should I use it?

Conversion-optimised cost per click. You nominate a conversion action and the platform optimises delivery towards users likely to complete it, while still charging per click. For a regional business it is the closest thing to geographic targeting available, because clean conversion signals teach the system to avoid areas that never convert. It requires the pixel or Conversions API, and enough monthly conversions to train on.

Will my B2B buyers actually see the ads?

Only if they are on the Free or Go tier and logged in. Plus, Pro, Business, Enterprise and Education users see no advertising. For B2B SaaS this is close to disqualifying. For most North East B2B sectors, where paid AI subscriptions remain uncommon in the SME base, the constraint is much weaker.

Should I run ChatGPT ads or work on organic AI visibility first?

Organic first, in almost every case. Paid placements reach only free-tier users, while organic citation reaches everyone including the paying subscribers who tend to be the more engaged buyers. Ad spend stops the day you stop paying. Citation authority compounds.

Where to start

If you want to know whether your business is currently recommended by ChatGPT, Gemini, Perplexity or Claude when someone in your sector asks for a supplier, that is measurable and it takes us a couple of days to establish. It is also the correct first question, ahead of any conversation about ad budgets.

We work with businesses across construction, engineering, professional services, hospitality and renewables from our base in North Shields, and we would rather tell you not to spend money on a channel than take a management fee for running it badly. You can see how that plays out across our work and case studies.

To talk it through, get in touch with our friendly team or call 0191 259 1234.