AI Search & GEO

ChatGPT ads are now cheap enough for a small business. That’s the wrong question.

The barrier to entry has collapsed — no minimum spend, about $25 a day, self-serve. But the published performance numbers disagree with each other by a factor of six, and nobody, including OpenAI, has a benchmark to give you. Here’s how to read the mess, and what I’d actually do with the money.

Key takeaways
  • ChatGPT ads now have no minimum commitment and a reported minimum daily budget of about US$25 per campaign.
  • OpenAI’s own help documentation states there are no performance benchmarks yet, so every figure you read is one vendor’s sample.
  • Reported click-through rates range from roughly 0.5% to 4% depending on who is publishing, and the people selling ChatGPT ad management report the highest numbers.
  • One agency measured a 0.21% conversion rate on ChatGPT ads against 3.71% on Google Ads for the same advertiser.
  • Only around 4% of AI citations about a brand point at that brand’s own website, which is why the paid question is the smaller one.
  • For most Australian service businesses the honest answer is still no — but for a specific reason, with specific conditions that would change it.

Earlier this year, advertising inside ChatGPT required a commitment somewhere around US$200,000. Today you can open an account, set a budget of about twenty-five dollars a day and be live by lunchtime. More than a hundred thousand advertisers have now run campaigns, across 89 countries.

So the question has changed. It is no longer “can I afford to test this.” It is “can I tell whether the test worked.” And that is a much harder problem than it sounds.

What it actually costs

There is no published rate card. OpenAI does not have one. You set a budget and bid, on either cost per thousand impressions or cost per click, in a self-serve auction.

Metric Reported range Who says so
Minimum commitment None OpenAI self-serve
Minimum daily budget About US$25 per campaign Third-party reporting; not published by OpenAI
Cost per thousand impressions US$16 to US$65 depending on source and category Agency samples; no official figure
Recommended opening bid US$3 to US$5 per click OpenAI guidance
Click-through rate 0.5% to 4%, depending entirely on who published it Similarweb, Graphite, various agencies
Official benchmark for any of this None exists OpenAI help documentation

Look at that click-through row again, because it is the whole article. Similarweb puts it near 0.68%. A Graphite panel reported 0.73% per ad-bearing reply. Meanwhile several agencies that sell ChatGPT ad management are publishing blended rates of 3.8% to 4%.

That is not a rounding difference. That is a sixfold gap on the single number that determines whether the channel is viable. And the pattern of who reports what is not subtle.

When the only people reporting good numbers are the people selling the service, you are not looking at data. You are looking at marketing.

Why the numbers disagree, and how to read them

Three reasons, and understanding them is more useful than memorising any single figure.

1. The samples are tiny and self-selected

An agency publishing a 4% click-through rate is reporting on the handful of campaigns it chose to run, in categories it chose to enter, for clients it chose to accept. Agencies that tried it, got nothing, and quietly stopped do not publish a blog post about it. You are seeing the survivors.

2. “Click-through rate” is being measured against different denominators

Some figures are clicks per ad impression. Others are clicks per reply that contained an ad. Others appear to be clicks per session. These produce very different numbers from identical underlying performance, and almost nobody states which one they are using.

3. The category spread is enormous

Developer tools and AI software report some of the strongest engagement, which makes sense — the audience is already inside an AI product, in a solve-this-problem frame of mind. A renovations company in Noosa is not that audience. Averages across all categories tell a local service business almost nothing about its own likely result.

The number that should worry you most

Click-through rate is the vanity metric here. Conversion is the one that pays. One agency reported running the same advertiser on both platforms and measuring a 0.21% conversion rate on ChatGPT against 3.71% on Google Ads — a seventeenfold difference.

A measurement firm that partners with OpenAI was blunter still, telling Modern Retail in October that the effective cost per click works out at around six times Google’s, that results have generally been poor, and that many of the sales being attributed to the channel do not look incremental.

That last word is the important one. Non-incremental means the customer was going to buy anyway. You paid for a sale you already had.

The maths for an actual small business

Forget the agency case studies with eight-figure budgets. Here is the arithmetic for a business with a $2,000 test budget, using the mid-range figures rather than the optimistic ones.

$2,000Test budget
~800Clicks at $2.50 each
~2Conversions at 0.21%
~84Conversions at 10.5%

Those last two columns are the problem in a single picture. Depending on which published conversion rate you believe, your $2,000 buys you either two leads or eighty-four. That is not a range you can plan against. It is a coin toss with your marketing budget.

And here is the part that makes it worse for a small business specifically: at two conversions, you cannot tell the difference between a channel that doesn’t work and a channel that works but needs a bigger sample. You have spent your entire test budget to learn nothing with any confidence. A big advertiser spending $200,000 gets a statistically meaningful answer. You get noise.

What the ChatGPT ads conversation keeps missing

Almost every article on this subject treats it as a media buying decision — what does it cost, what’s the return, should I shift budget. That framing quietly assumes the only way into an AI answer is to pay for it.

It isn’t, and the organic side is where the interesting numbers live.

When someone asks ChatGPT or Perplexity for a recommendation, the assistant assembles an answer from sources across the web and frequently names businesses inside the answer itself. That mention is free. It also sits above the ad, carries no “sponsored” label, and arrives with the assistant’s implied endorsement rather than against it.

Research published through 2026 suggests this is a genuinely open field. An analysis of more than 22,000 citations from AI search found that over a third pointed to domains with a Domain Authority under 40 — sites that would struggle to reach page one of Google for the same query. Separate work found that roughly 80% of the URLs cited by ChatGPT and Perplexity do not appear in Google’s top 100 results at all.

What that means in practice

The organic path into AI answers is not gated by the thing that gates Google. A small, specific, well-structured site can be cited alongside — or instead of — a competitor with twenty times its authority. The paid path, meanwhile, costs roughly what Google costs and converts, on the available evidence, considerably worse.

Spending on the paid door while the free one stands open is a strange allocation of a small budget.

Where I land, and what would change my mind

For an Australian service business — a trade, a practice, a consultancy, a local operator — I would not put money into ChatGPT ads this quarter. Not because the channel is bad, but because you cannot yet measure whether it worked, and a test you cannot measure is not a test. It is a donation.

That is a soft answer unless I tell you what would change it, so here are the specific triggers. These are the things I am actually watching.

I’d revisit this when any two of these are true
  • OpenAI publishes performance benchmarks by category, so you can judge your own results against something
  • Independent measurement partners are in place and reporting incrementality, not just attributed sales
  • Geo-targeting goes below country level, so an Australian business can stop paying for impressions in markets it cannot serve
  • Conversion figures from non-vendor sources converge into a range narrower than the current sixfold spread
  • Someone publishes a credible case study from a business of your size, in your category, in your country

Two caveats, because the argument is not one-sided.

If you sell software, developer tools or anything where the buyer is already inside an AI product when the need occurs to them, the case is meaningfully stronger than it is for a local service business, and the early engagement figures reflect that. And there is a real argument for buying early in any channel: costs rise as competition arrives, and the advertisers learning the mechanics now will be better at it in a year. I think that argument is weaker than usual here, because what you’d be learning is how to operate a measurement system that doesn’t exist yet. But it isn’t a silly position.

Being early to a channel is valuable. Being early to a channel you cannot measure is just being early to the bill.

What I’d do with the money instead

If you have a few thousand dollars and you want to be found by people using AI assistants, here is the order I would spend it in. None of this is speculative and all of it compounds.

Make sure the assistants can read you at all

A surprising number of sites block AI crawlers without realising, or render their key content in a way that machines cannot parse. This costs nothing to check and nothing to fix, and no amount of ad spend compensates for it.

Get your business details identical everywhere

Assistants resolve you to an entity before they can recommend you. Inconsistent details across your site, your Google Business Profile and third-party listings make that resolution harder, and an assistant that is not confident who you are will name someone it is confident about instead.

Earn third-party mentions, because that is where the citations actually are

This is the finding that most surprises people. One index of brand citations found that only about 4% pointed back at the brand’s own website; the other 96% were on review sites, community forums, editorial coverage and listings. Another analysis put earned media at the large majority of citations. Your own site matters — but the roundups, directories and industry listings that mention you matter more.

It is worth saying plainly that these studies disagree with each other too. At least one vendor analysis put brand-owned domains at over half of all citations, which is nearly the opposite finding. The honest summary is that the proportion is contested, but no serious study puts your own website anywhere near the whole picture.

Publish content that is specific enough to quote

Assistants cite passages containing concrete facts — numbers, named places, dated results, actual prices. Generic advice gets summarised and dropped. The content that earns a citation is the content that says something a model cannot assemble from everywhere else.

Frequently asked questions

How much do ChatGPT ads cost in Australia?

There is no published rate card. There is no minimum commitment and a reported minimum daily budget of around US$25 per campaign. Reported costs sit between roughly US$16 and US$65 per thousand impressions depending on category and source, with OpenAI recommending opening bids of US$3 to US$5 per click. Billing is in your account currency. Geo-targeting is country-level only, so you can target Australia but not a specific city or postcode.

Do ChatGPT ads actually work?

Nobody can answer this reliably yet, including OpenAI, whose own documentation states there are no performance benchmarks across advertisers, industries or campaign types. Published results vary enormously: one agency reported a 0.21% conversion rate against 3.71% on Google Ads for the same advertiser, while vendors selling ChatGPT ad management report far stronger figures. A measurement firm partnering with OpenAI reported in October 2026 that results have generally been poor and that many attributed sales appear non-incremental, meaning the customer would have bought anyway.

Is it better to advertise in ChatGPT or to be recommended organically?

For most small businesses, organic. An organic mention appears inside the answer itself rather than beside it, carries no sponsored label, and costs nothing per appearance. The available research also suggests organic AI citation is unusually accessible to smaller sites: one analysis of over 22,000 citations found more than a third pointed to domains with a Domain Authority under 40, and separate work found around 80% of URLs cited by ChatGPT and Perplexity do not appear in Google’s top 100 results.

Who actually sees ChatGPT ads?

Logged-in adult users on the free and lower-cost plans. Subscribers on the higher-tier paid plans do not see ads at all, and users under 18 are excluded. OpenAI also blocks ads from appearing alongside health, mental health and political topics. For businesses whose customers skew towards paid AI subscriptions, this removes a significant portion of the addressable audience before any targeting decisions are made.

What should a small business do about AI search right now?

Check that AI crawlers can access and parse your website, make your business details identical across your site, your Google Business Profile and third-party listings, pursue mentions on the review sites, directories and industry roundups that assistants actually cite, and publish content specific enough to be quoted rather than summarised. All of this is free or cheap, compounds over time, and directly addresses how assistants assemble recommendations. Paid placement can wait until it can be measured.

Sources

Not sure whether AI assistants can even see your business?

That’s the question worth answering before you spend anything on ads. I run AI visibility audits for Australian service businesses — what the assistants currently say about you, where your competitors are being cited instead, and what to fix first.

Book a free 15-minute call

Similar Posts