Marketing: ad platforms opened campaigns to outside AI agents

Marketing's ad-buying controls are opening to outside AI agents: as of mid-2026, the largest ad platforms let a general assistant such as ChatGPT plan, build and adjust a paid campaign directly, without the marketer opening the platform's own dashboard. On 18 June 2026 Snapchat opened its ad system to third-party AI agents, and Meta did the same this spring. The craft a small agency has sold for years, running an ad account well, is becoming something an AI can drive.

The buying side opened up

When Era Haus last looked at this industry in AI is rebuilding how you get found and buy ads, the change was that Google had folded campaign management into its own AI assistant and opened ChatGPT's ad system to small budgets. The work of the media buyer, the person who plans and runs paid ads, was moving into software. A month on, that software stopped being walled inside each platform.

On 18 June 2026 Snapchat opened its advertising system to outside AI agents through an MCP server, a standard connection that lets an AI assistant read and change a live campaign on its own (Adage, June 2026). The same day it launched Snap Smart Assistant, which proposes campaign goals, audiences and settings in plain conversation. Meta opened the same kind of access on 29 April, letting an outside assistant handle budgets, reporting and campaign edits, and through the spring it lowered the bar for third-party tools to plug in (Digiday, June 2026).

The shift for an operator is concrete. You no longer run a campaign only inside the platform's own screens; you can sit in one assistant and have it adjust spending and creative across accounts. The setup-and-tuning that justified a monthly retainer is becoming something a client could point an AI at themselves.

The discovery side kept moving

The other half of the May picture, how customers find you, also moved. For two decades being found meant ranking high in a list of links. Now a growing share of people take an answer straight from an AI that may never send them to your site. Google's AI Overviews, the AI answer placed above the links, now show on about a quarter of US searches, and assistants like ChatGPT answer inside their own apps (Similarweb, June 2026).

The early traffic data is clearer than it was a month ago, and it cuts both ways. AI assistants still send only about 1% of all website visits, so this is not yet where your customers come from. But that traffic converts unusually well, second only to paid search in one 2026 analysis, and it rose sharply in early May when ChatGPT began turning brand names in its answers into clickable links (Goodie AI search report, 2026). Being named inside the AI's answer is becoming the thing worth optimising for, more than ranking in a list of links. Marketers have a name for it: answer engine optimisation, getting cited in the AI's reply rather than ranked beneath it.

A new rule on the creative itself

There is also a new rule landing on the ads you make, and it is clearest in Europe. From 2 August 2026 the EU AI Act requires that AI-generated or AI-altered ad content, whether image, audio, video or text, be marked as artificially generated, with machine-readable provenance attached (European Commission, 2026). Getting transparency wrong can cost up to 15 million euros or 3% of worldwide turnover. The United States has no equivalent federal rule; the US Federal Trade Commission has warned about deceptive AI content but set no blanket labelling requirement. If you run AI-made creative for customers in the EU, the labelling duty arrives in months and is already worth preparing for.

What it means for you

For a small marketing team or agency, two of the things you sell are being squeezed at once. Running ad accounts well is becoming a job an AI assistant can do from outside the platform, on a budget a client could manage alone. And ranking a website, the core of old search work, matters less when the customer takes an answer from an AI that may never name you.

That points to where the value actually sits, and it is the argument Era Haus made in defensibility in the AI era: when a skill becomes cheap and common, worth moves to what does not copy. In marketing that is a clear point of view, original research or data an AI has to cite, the judgement of what to say and to whom, and a client's trust that a person stands behind the result. The agency that only ran an ad account is exposed. The one that decides what a business stands for, and becomes a source the AI quotes, keeps its value.

What to do about it

Three grounded moves. First, find out whether AI answers name you at all. Ask ChatGPT, Google's AI answer mode and a couple of rivals the questions your customers actually ask, such as the best provider of your service in your city, and see who gets cited. If it is not you, clear, factual, well-structured pages an AI can quote are how you earn a place in the reply.

Second, treat the new agent-run ad tools as a way to do the work faster, not as a reason to hand over the account. Let an assistant draft and adjust campaigns, but keep a person approving what it spends. An AI optimising toward a number it was given can quietly push budget toward the wrong goal.

Third, if any of your creative is AI-generated and runs in Europe, start labelling it and keeping the provenance data now, ahead of the August deadline. Building the habit while it is optional is cheaper than retrofitting it under a 15-million-euro rule.

What to watch rather than act on yet: letting an outside agent run a campaign end to end with no one checking. The access is now there on the major platforms, but the tools are new and nobody has trustworthy benchmarks. Put a small, capped budget through one, measure it against your other channels, and scale only if it holds.

The pattern underneath

The shape is the one this series has traced across law, property and accounting: AI gets cheap inside the tools a profession already uses, the routine craft is automated, and value moves to judgement and relationships that do not copy. Marketing's version is sharper, because it sits on top of the very thing AI is moving fastest, how people find and decide what to buy. The team using these tools to do the old job faster gets the small win. The one working out how to stay the answer a customer trusts, now that the answer increasingly comes from a machine, is working on the real one.