ChatGPT Ads vs Meta Ads vs Google Ads: targeting

Google, Meta and ChatGPT all sell ad space assigned by an artificial intelligence, but the signal each one starts from is different. Google starts from the search someone has just typed. Meta starts from the sequence of things that person watched, clicked and ignored in the preceding weeks. ChatGPT starts from the conversation open at that moment.

The shift is not about where we buy media. It is about which intent signal we are trying to catch, and today the three platforms read that signal in ways that do not overlap.

What you should not do is turn the difference into a caricature. Google does not run on keywords alone, Meta does not run on interests alone, ChatGPT does not run on chat text alone. All three combine AI, first-party data, contextual signals, creative, landing page and conversions. But the dominant starting point stays different, and that is where the job changes.

	Google Ads	Meta Ads	ChatGPT Ads
Characteristic signal	search and query	behaviour and likelihood of interest	conversation and contextual intent
Is the person searching?	often yes	not necessarily	trying to understand or solve
What you describe well	demand, offer, landing, conversion	offer, creative, constraints, value signals	situations, needs, context, message
Where it can be strong	demand that already exists	discovery and demand generation	problems that need comparison
Typical mistake	still thinking in keywords only	still thinking in interests only	treating context hints as keywords

The table names the most characteristic signal of each ecosystem, not the only one it uses.

Why this comparison matters now

ChatGPT Ads is no longer an American test. Since 31 August 2026 self-serve access from OpenAI's panel has been open across Europe, inside the expansion announced on 18 August and also reported by Digiday in OpenAI's ChatGPT ads business hits $1 billion run rate. The country list changes without announcement, so the only reliable reference is the panel's availability page on the day you check.

Meanwhile Google is pushing AI Max into Search campaigns and Meta keeps moving ranking and recommendation towards models that read longer and longer behavioural sequences. All three head in the same general direction, more AI and fewer rigid manual rules, but they get there from different information. That is where the advertiser's craft changes.

Google Ads: catching demand that already exists

Google Search starts from one of the clearest intent signals in existence, which is somebody typing what they are looking for.

Take "PVC window quote". We do not know that person's budget, we do not know how close to buying they are, we do not know whether they will pick us. But we know the thing that matters most: at that moment they are expressing a precise need, already turned into words. That is Google's historic strength, and it has not gone anywhere.

Google is no longer "I write the keyword and decide everything"

This is the part most often described with an outdated view of the panel. Keywords still exist, exact, phrase and broad match still exist. But matching has become far more semantic, and the defaults have switched sides.

Broad match is assigned by default to all your keywords, and the keyword matching options state that with it ads can serve on searches "that don't contain the direct meaning of your keywords". The keyword stopped being a condition and became a hint.

On top of that sits AI Max, which is not a new campaign type but an optimisation layer inside existing Search campaigns. When it is on, how AI Max campaigns work describes extended matching that combines broad match, keywordless technology, learning from the keywords, creatives and URLs already in the campaign, and signals processed in real time.

The detail that matters more than the definition: AI Max is on by default in new Search campaigns. Open one today, touch nothing, and you have it.

So Google is still the place where declared intent carries enormous weight. But it is no longer correct to picture it as the system where you decide word by word everything that can trigger an ad. The question moved from "which keyword should I buy" to "which demand am I trying to catch, and what signals am I giving Google so it recognises it".

Meta Ads: catching interest before it becomes a search

On Instagram and Facebook the context is inverted. Nobody opens the app typing "I want to buy a sofa". They open it to watch things, they scroll, they stop on a Reel, they skip a post, they visit a profile, they click an ad, they come back two days later.

So Meta has to answer a harder question than Google's: which ad might interest this person, who right now is not looking for anything. That is why it has always been formidable at discovery. It does not only intercept existing demand, it can help create it.

Meta does not only look at what you like

The interesting part of today's systems is that behaviour is not read as a list of labels. Meta has been working on sequence models for years: in 2024 it published Sequence learning: a paradigm shift for personalized ads recommendations, showing that modelling the order and timing of user actions produces richer representations than hand-engineered features, and in 2026 it described From User Sequences to Scaling Laws, the multi-stage architecture running on top. Not just "clicked a running post", but in which order the actions arrived and how much time sat between them.

The result is a representation of interests far richer than any static hand-picked list.

That does not mean manual targeting is dead, and whoever repeats that line has usually never tried switching off exclusions on a serious account. Location, age, custom audiences, exclusions, first-party data and constraints still matter. What changed is the weight: we used to describe the audience precisely, now the advantage sits in the inputs, meaning a good offer, genuinely different creatives, reliable conversion signals, first-party data, the right guardrails and enough volume for the system to learn.

Less obsession with micro-targeting, more quality in what you feed it. Which controls are still worth keeping by hand, and which are worth leaving to the automation, is the whole argument of what's left for you to do.

ChatGPT Ads: when the signal is the problem you are describing

Here the kind of available information changes, not just the amount.

ChatGPT Ads starts inside a conversation. OpenAI's documentation on how ads get selected explains that the system can consider, among other signals, the context and intent of the current conversation, the landing page, the ad headline and copy, the context hints the advertiser provides, the targeting settings and, when ad personalisation is on, selected signals from the person's wider ChatGPT experience.

A query tells you what someone is looking for. A behavioural sequence helps estimate what might interest them. A conversation can contain the problem, the goal, the constraints, the doubts and the alternatives all at once.

The same prospect seen by all three platforms

Say I sell Meta Ads consulting.

On Google that person searches "Meta Ads consultant". Three words, crystal clear signal, they are looking for a solution.

On Meta they search nothing, but they follow marketing content, engage with other people's advertising, visit business pages, click lead generation posts. Meta tries to work out whether my ad could be relevant before the explicit search exists.

Inside ChatGPT they write something like this:

I'm spending €3,000 a month on Meta but I get a lot of leads who never answer. CPL is low, yet commercially I close nothing. What should I be checking?

They did not write "I want a consultant". They described that they use Meta Ads, how much they are spending, what the problem is, which metric they are watching and why the result disappoints them. In advertising terms that is an extremely rich context, and none of it existed in the three words of the search.

An ad about lead quality or tracking can therefore be relevant to the situation described, not because the user typed an exact commercial keyword. That is the real novelty of conversational advertising.

Context hints are not keywords in disguise

This is the most important thing to grasp when you arrive at ChatGPT Ads with a Google Ads head.

At ad group level you can supply context hints. They help the system understand what the product offers, who it helps, in which situations it is useful and which conversations are relevant. But they are not exact-match keywords and they do not guarantee your ad appears in any given conversation.

Writing "Meta Ads consultant" as your only context hint means using a new tool with old logic. Describing the moment works much better:

Companies investing in Meta Ads who receive poorly qualified leads, have an apparently good CPL but low contact or sale rates, and are trying to work out where the process breaks.

The difference looks subtle and is enormous. You are describing a situation, not a word. How the panel is built, who sees the ads and how measurement is set up is all in the note on how ChatGPT Ads works.

Query, behaviour and conversation: three kinds of intent

The fastest way to keep them apart is to name the signal, not the platform.

Search intent is what a person declares by typing. "Hotel Rome centre" expresses the need but almost none of the context: you do not know if they travel for work, with kids, tomorrow or in six months.

Behavioural intent is what a platform infers. Nobody wrote "I want a bike", but they have been watching cycling content for weeks and visiting compatible pages. Meta builds a probability, not a certainty, and sometimes it is embarrassingly wrong.

Conversational intent is what surfaces while someone tries to understand something. For instance: "I'd like a bike to commute, I do 18 km a day, I live in a hilly city, I don't want to arrive sweaty and I have about €2,000". There is no "cycling" interest here. There is use, distance, problem, environment, desire and budget in one sentence.

Two people identical in age, city, revenue and interests can be commercial opposites: one is hiring, the other is selling the company. Demographics do not tell them apart. A conversation does.

Not three alternatives, three moments

The useful question is not which platform will replace the others, but at which point in the decision each one can catch your customer.

One possible path: on Meta I discover a problem or a product I was not looking for, inside ChatGPT I try to understand it and compare options, on Google I search for a brand, a price, a review or a specific supplier, then I convert on the site.

But the order shuffles constantly. I can start on Google, ask ChatGPT for a comparison, get retargeted on Instagram and buy directly. The point is that demand no longer lives in one place, and assigning the whole value of a sale to a single step by reading one platform's panel makes less sense every month.

How campaign structure changes

The practical consequence is that I would not group ads the same way on the three platforms. Not because a mandatory universal structure exists, but because the kind of intent I am organising changes.

On Google I think in demand clusters. For a CRM: CRM for small companies, CRM for sales teams, lead management software, CRM with WhatsApp, alternatives to Excel for sales. Different ways of expressing the same request.

On Meta I think in motivations and creative angles. Chaos in lead handling, lost opportunities, control over the sales team, automation, reporting, speed of follow-up. Here the creative does not present the CRM, it gives Meta different reasons to find relevance.

On ChatGPT I think in situations. Groups built around conversations like "we get plenty of contacts from campaigns but nobody manages to call them back in time", "every rep uses their own file and we have no shared pipeline", "we'd like to automate the follow-up once a lead arrives", "we can't tell which campaigns produce actual customers and which only produce contacts".

Those are not four targets in the traditional sense. They are four moments of need, and you recognise them because you have heard them on the phone.

The same offer written for three platforms

Say we sell a CRM for small businesses.

On Google the headline is "CRM for SMEs: manage leads and sales". I am catching demand that already exists, and the best thing I can do is be recognisable.

On Meta the hook is "how many leads do you pay for every month that nobody ever calls back?". I make a problem visible and create interest before the person actively searches for a CRM.

Inside ChatGPT the message is "getting leads from Meta and Google but follow-ups still live in Excel? Centralise contacts, pipeline and sales activity in one CRM". I step into a situation the person has already articulated.

Same product, three ways of building relevance.

ChatGPT Ads does not mean paying to become the answer

Worth stating plainly, because it is the commonest misunderstanding. Ads inside ChatGPT are shown separately from the assistant's answers, and advertising does not determine what ChatGPT says.

The model is not "I pay and ChatGPT recommends my product". It is: the conversation creates a context, the system checks whether an eligible ad exists, the ad appears below the answer and not inside it.

There is a second misunderstanding on the opposite side. The fact that the system uses conversation context does not mean the advertiser receives the chat: advertisers get no conversations, no memory and no personal data about the person writing. They get campaign data. The context helps the system decide relevance, it does not become a dossier handed to the brand.

ChatGPT Ads has traditional levers too

Conversational advertising does not mean the classic controls are gone. The options described in Campaign Targeting let you work on country and, where supported, region, city, metro area and postcode, then on platform and on the custom audiences you include or exclude. Custom audiences are built from your own customer or prospect lists using the supported identifiers, and are limited to eligible accounts.

Conversation is the distinctive signal, not the only one. Like Google and Meta, this system is hybrid too: context plus audience plus creative plus offer plus conversion signal.

Targeting matters, the conversion tells the system what counts

All three platforms share the same underlying problem. They are excellent at finding people who click, fill in forms, visit and engage. But if the signal you send back is mediocre, the algorithm will optimise flawlessly towards the wrong thing.

A hundred leads at €8 look better than fifty leads at €15. If two of the first hundred become customers and ten of the second fifty do, CPL was telling you an incomplete story with perfectly correct numbers.

That is why tracking quality stays central everywhere. On ChatGPT conversions go through OpenAI's pixel, the Conversions API or both, and when the same conversion comes from both you need the same event identifier so the system can drop the duplicate. Meta has its own Conversions API, Google works with conversion measurement, enhanced conversions and offline conversion imports.

The names change, the principle does not: the more the system automates finding people, the more precise you have to be about which results have value.

Why you cannot add up conversions across the three panels

Lining up the numbers is not enough. Each platform can use different attribution windows, different attribution methods, modelled conversions, its own deduplication logic, and timestamps and time zones that do not line up.

So "Meta says 50, Google 30 and ChatGPT 10, I made 90 sales" does not hold. The final economic source is your ecommerce, CRM or accounting system. The panels tell you how each platform attributes its own contribution to itself; the business tells you how many customers you actually generated.

Which platform to choose

There is no universal answer, but there is a question to start from: where does demand for my product start today?

Google is strongest when demand already exists, people can describe what they want, commercial intent is clear and the need is urgent or specific. "Dentist Milan open Saturday" is the textbook case.

Meta is strongest when you need discovery, the product is visual, the problem is not being searched for yet, good creative can make a need obvious and you have enough volume for the system to learn.

ChatGPT gets interesting when the problem needs explaining, alternatives are many, the decision runs through a comparison and the value of the offer emerges from context. Software, professional services, training, technology, B2B, travel, complex products, home solutions. Not because it will automatically work better in those sectors, but because there the conversation carries more useful information than a short query.

There is one filter to apply before all the others, though, and it is the only one that can zero out the channel: ads are shown only on the free and Go plans. Anyone paying for Plus, Pro or a business plan never sees them. If you sell to professionals who live inside ChatGPT all day, a large part of your audience is on a subscription and this channel does not reach them, however rich the conversation is.

What changes for people doing performance marketing

The interesting part is not one more panel. It is where the value of the work moves.

For years much of the advantage sat in knowing the right keyword, the right interest, the right Lookalike, the right structure, the hidden setting. Those skills do not disappear. But as all three platforms automate matching, bidding, ranking and delivery, other things become worth more.

Understanding the problem instead of describing the target. Building a strong offer, because no algorithm makes something interesting that the market does not want. Writing genuinely different messages, not thirty cosmetic variants of the same ad. Supplying quality signals, meaning qualified leads, sales and reliable events. Measuring beyond the click, because CTR and CPC can improve while the business gets worse. And understanding people, because platforms change constantly and human motivations change far less.

The real evolution of targeting

So where does targeting really change between Google Ads, Meta Ads and ChatGPT Ads? Google mostly catches what a person is searching for. Meta mostly tries to predict what might interest them. ChatGPT can catch what they are trying to understand or solve, while they describe it.

None of those three sentences fully describes the platform it refers to. But each describes the moment of intent that ecosystem starts from, which is the thing you need to know when deciding where a budget goes.

The value of conversational advertising is not in being the new Google and not in replacing Meta. It is in having brought advertising into a moment that used to be almost unreadable: the one where a person is putting their problem into words.

Targeting is not disappearing. It looks less and less like a list of filters and more and more like a comprehension problem. Which means the competitive advantage, for advertisers, will be less about knowing a button nobody else knows and more about understanding the business, the problem, the person, the message and the signal to send back to the algorithm.

Frequently asked questions

What is the main difference between ChatGPT Ads and Google Ads? Google Search starts mostly from a query and search intent. ChatGPT Ads can use the context and intent of the current conversation. Context hints help matching but do not work as exact-match keywords.

Are ChatGPT Ads context hints keywords? No. They can contain themes, needs, situations and relevant words, but they are not exact-match and do not guarantee your ad runs when someone uses a given phrase.

Do ads influence ChatGPT's answers? No. Ads are separate from the assistant's answers and do not determine what ChatGPT says. Advertisers receive no conversations, no memory and no personal data about the person writing.

Can I reach anyone on ChatGPT, the way I can on Google and Meta? No, and it is the difference that weighs most on the choice. Ads inside ChatGPT are shown only on the free and Go plans: anyone paying for Plus, Pro or a business plan is out of reach. Google and Meta have no equivalent filter.

Does Meta Ads still work with interests and audiences? Yes. Manual targeting has not disappeared. But Meta's ranking systems use far richer behavioural sequences, so in most strategies the advantage shifts to creative, conversion signals and data quality.

Does Google Ads still use keywords? Yes. Search campaigns still use them. But broad match is the default and can serve on searches that do not contain the direct meaning of the keyword, and AI Max increases the weight of semantic signals, creatives and URLs.

Does AI Max replace Search campaigns? No. It is an optimisation layer inside existing campaigns, not a campaign type. It is on by default in new campaigns, and its settings can be managed.

Can I compare ROAS across the three platforms directly? Carefully. Models and attribution windows differ. To see the real economic outcome, compare advertising data against ecommerce, CRM or accounting, and where possible run incrementality tests.

Which should I choose between Google Ads, Meta Ads and ChatGPT Ads? It depends where demand sits. If it is already expressed in a search, Google. If it has to be created, Meta. If the person is trying to understand or compare, ChatGPT adds a touchpoint that did not exist before. In most cases the answer is using them with different roles, not picking one.

ChatGPT Ads, Meta Ads and Google Ads: where targeting really changes