70% of Ad Spend Will Pass Through Machine-Made Decisions

Gartner expects more than 70% of global ad spend to flow through AI-influenced self-serve platforms by 2028.¹ As platforms automate audience selection, product placement, campaign adjustment and measurement, marketers are gaining speed while delegating decisions they once made more visibly.

The most consequential marketing launches this week were not really about making another advert faster. ChatGPT Ads expanded, customer data moved closer to conversational interfaces, product feeds were prepared for shopping agents, and creator programmes became manageable at a scale that once needed a team. The common thread is simple: the system is starting to act on the brand’s behalf, not merely produce material for someone else to approve.

The interface now makes decisions

ChatGPT Ads launched in five additional markets this week, while Decile introduced an ecommerce analytics and activation MCP designed to connect enterprise AI tools directly to enriched first-party customer data.²³ Put those developments together and the marketing interface starts to look very different. A marketer can ask a question about customers, turn the answer into an audience and operate closer to the same conversational environment where paid distribution is also appearing.

That collapses several steps that used to live in separate products. Research happened in one tool, customer analysis in another, audience creation somewhere else, campaign setup in an ad platform and reporting in a dashboard after the fact. The new model compresses those handoffs, which is useful, but it also means a poor assumption can travel further before anyone notices it. Fewer interfaces do not mean fewer decisions.

This is where first-party data becomes more valuable than another clever prompt. Generic systems can suggest segments and campaign ideas, but a system connected to actual customer behaviour can work from what people bought, returned, ignored, repeated or abandoned. That gives AI in marketing something concrete to reason over, and it changes the quality of the recommendation before creative production even starts.

The practical implication is uncomfortable for teams that have spent the last year measuring adoption by the number of people using generative tools. The strategic question is now whether the underlying customer data is clean enough, permissioned enough and understandable enough to support automated decisions. A conversational interface can make analysis feel easier while hiding the fact that the inputs are fragmented, badly labelled or commercially irrelevant.

Machines need clean facts

Productsup offered a second clue. The company, which says it processes more than two trillion products a month across more than 2,500 channels, added Glance as a native export destination so brands can map existing product feeds into an agentic commerce channel.⁴ The important work happens before the recommendation: attributes need to be complete, products need to be described consistently, and the system needs enough structured information to match an item to a shopper’s stated intent.

That sounds operational until an agent chooses one brand over another. At that point, catalogue quality becomes part of brand visibility. A beautifully written landing page cannot compensate for contradictory prices, missing product attributes or inconsistent naming across a website, marketplace feed and third-party listing.

Semantic XEO is attacking the same issue from a different direction, with a framework for structuring company knowledge, evidence and relationships so systems such as ChatGPT, Claude, Gemini and Perplexity can retrieve and verify it.⁵ The terminology will continue to evolve, but the marketing discipline underneath it is familiar: say the same true thing clearly wherever customers and machines are likely to look. The difference is that a machine can compare those sources at a scale no customer ever could.

This matters for Instagram content strategy too. A business using an Instagram marketing AI tool still needs accurate source material, recognisable language and a stable account of what it sells before the system can produce useful posts. Asteris approaches AI-powered Instagram content from that direction, using the business’s own media and website context rather than treating a blank prompt as the starting point. The same principle applies far beyond social content: better outputs begin with better evidence.

Distribution is turning into a default

Amazon made the delegation issue more visible by moving Sponsored Products into creator content and automatically enrolling existing campaigns, with current bids and budgets carrying over.⁶ Advertisers who did nothing could therefore begin paying for a new placement type. That is a small interface choice with a large operational consequence because the platform default starts making a media decision before the marketer has actively made one.

This is likely to become normal. Platforms have every incentive to reduce setup friction, increase inventory utilisation and let automated systems shift spend towards whatever placement appears most likely to perform. Gartner’s forecast that more than 70% of global ad spend will flow through AI-influenced self-serve platforms by 2028 describes this direction at scale.¹

The problem for a brand is not automation itself. Automated buying can remove hours of repetitive configuration and can respond faster than a person watching a campaign manually. The risk appears when a team stops knowing which choices the system has permission to make, what changed between one reporting period and the next, and whether a performance lift came from better marketing or from a newly activated placement the team never evaluated.

Creator marketing shows the upside when the boundaries are clearer. Turo says one person managed 435 creator partnerships in a year using Agentio, with the platform handling work such as matching, pricing, contracting and performance tracking.⁷ That is a strong example of automation making human judgement more scalable because the marketer can spend less time coordinating paperwork and more time deciding which creators fit, what good work looks like and how the programme should evolve.

There is a second-order consequence to that efficiency. When coordination cost falls, the ceiling on programme size rises, so a brand can work with more niche creators without adding equivalent headcount. That could broaden participation if marketers use the extra capacity to find smaller creators with stronger audience fit, rather than using automation only to push more spend through the same familiar names.

The marketer’s role changes with it. Someone still has to recognise whether a creator’s humour fits the brand, whether an audience relationship feels earned, and whether the brief leaves enough room for the creator to sound like themselves. Automation is valuable precisely because it can remove the repetitive administration around those judgements without pretending the judgements disappeared.

Cheap output raises the bar

LinkedIn’s new option for reporting content that “seems like AI slop” landed in the same week that tools continued promising publish-ready video and advert production from minimal inputs.⁸ The juxtaposition is useful. Platforms want more creation because more content gives people more reasons to use their products, but feeds lose value when the extra supply becomes indistinguishable.

The production bottleneck is disappearing faster than the judgement bottleneck. A product page can become a video, one campaign can become dozens of variants and a small team can schedule more content than it could have produced manually two years ago. None of that creates an opinion, a customer insight, an earned visual detail or a reason for a particular audience to recognise the work as belonging to this brand.

Google’s Gemini campaign with Shah Rukh Khan makes the alternative visible. The campaign focused on multilingual use beyond India’s metros and used a cultural figure whose relevance is part of the message, rather than asking the audience to admire the technology in isolation.⁹ P&G’s Always Discreet campaign with mothers of NFL players and Bazooka’s use of MomTok rivalry point in the same direction from entirely different categories: the valuable choice is often who, where and why, before anyone asks how many assets can be generated.

For teams wondering how to stay on brand with AI content, the answer starts before the model writes anything. Give the system real photographs, product facts, customer language, previous examples and explicit constraints about what the brand would never say. For Instagram AI content management in particular, those inputs matter because a consistent publishing cadence can magnify weak source material just as efficiently as it can magnify strong material.

Measurement has to survive the platform

More delegated decisions also make independent measurement more valuable. Gravity’s agent-to-agent advertising model is designed to supply commercial product information directly to shopping bots, which means a paid influence can shape a shortlist before a person encounters a conventional advert.¹⁰ That may be useful for discovery, but it complicates the customer’s ability to see where recommendation ends and commercial influence begins.

The same week, reporting around manipulative ad fraud described hidden activity designed to manufacture clicks and engagement signals at scale.¹¹ A platform can optimise brilliantly against a signal that has little relationship to human interest if the signal itself is contaminated. That makes more automation dangerous when measurement quality does not improve with it.

The useful response is not to retreat into manual reporting. It is to decide which evidence the organisation trusts and to keep some of that evidence outside the platform whose performance is being judged. Revenue, repeat purchase, qualified pipeline, customer retention and direct feedback may each matter more than another platform-specific score, depending on the business.

Events can also become part of that evidence. RainFocus has increasingly positioned its platform around using attendee behaviour and AI to connect events with sales and revenue data, giving marketers a way to relate what happened in the room to what happened afterwards.¹²The better loop is behaviour back into judgement, not content back into more content.

Better inputs become the advantage

The next phase of AI content marketing will reward teams that know what their systems are allowed to decide and what those systems are allowed to learn from. More of the execution layer can be compressed, from creator coordination to audience building to product-feed distribution. That makes the quality of source material, customer data, permissions and measurement more visible because weak inputs can now travel through the organisation faster.

This should be good news for marketers who know their customers and products well. The work that becomes easier is the repetitive coordination around the decision, while the decision itself still needs context: which segment matters, which creator belongs, which product claim is defensible, which placement is acceptable and which result deserves belief. Those are commercial judgements, not formatting tasks.

The companies that benefit most will not be the ones that connect every new tool first. They will be the ones that can explain where their customer truth lives, which facts machines may use, how brand voice is preserved and how a claimed performance gain will be checked. That is a much more demanding operating model than prompt-and-publish, but it is also a more useful one.

If 70% of ad spend really does pass through AI-influenced platforms within two years, marketers will spend less time moving the machinery by hand. They will spend more time deciding what the machinery should be trusted to do. That is a better use of human attention, provided the business has done the unglamorous work of giving the system something reliable to act on.

Sources