By Alex Springer, Director at OpenAttribution.org
We are long overdue a change in how we value the way content and knowledge is produced, shared and sold. It is time to move past legacy media metrics such as impressions and clicks, and to be more precise about what attention can and cannot tell us.
Attention is not a flaw in itself. It is a necessary ingredient for an ad exposure to have any impact. The problem comes when impressions are treated as a reliable proxy for attention, particularly when they are bundled with privately controlled attribution systems that claim to explain what happened next.
These measures are vulnerable to exploitation. Invalid ad traffic alone is projected to cost advertisers more than $172 billion by 2028. The programmatic industry’s dependence on measurable delivery has created incentives to manufacture the appearance of attention at scale, whether through made-for-advertising sites, non-human traffic or large volumes of low-quality content.
Programmatic technology did more than change the way advertising was bought and sold. It also changed the underlying good being transacted. Publisher relationships with readers were flattened into units of inventory, while advertisers were asked to trust a complex supply chain and attribution systems they could rarely inspect.
Opaque programmatic stack
Martech—and its emphasis on first-party data—offers some relief because the advertiser has a direct relationship with its audience. The company has a better understanding of whom it is talking to, and its investment goes into direct communications rather than evaporating into an opaque programmatic stack.
Add Customer Data Platforms that fill in some of what customers do elsewhere, together with social platforms that straddle adtech and martech, and the result is a powerful, occasionally frightening, audience profiling and targeting system.
Had social and search platforms stayed in their lanes, the traditional exchange between content producers, advertisers and audiences might have remained viable for all three. Instead, these platforms spent years finding ways to keep users inside their own environments, often by reformatting third-party content through Google snippets, AMP and Facebook Instant Articles.
Consumer generative AI extends that approach. Chatbots, AI search, personal assistants and “thinking partners” offer a more complete way to answer a user’s question without sending them to the sites from which the underlying information came. Content remains much cheaper to scrape than to create, while the resulting product gives the platform a stronger hold over the user relationship.
AI has also put pressure on a particular part of the publishing market: sites built largely on aggregating, rewriting and redistributing information available elsewhere. Generative AI can do much of that work more cheaply than a publisher ever could. Publishers using LLMs to increase their output of similar material are likely to capture diminishing value as the supply continues to grow.
What remains is a smaller, perhaps more appropriately sized, publisher market based on original work, human connection, brand trust and topic authority. The audience relationships involved may be fewer, but they are also likely to be deeper.
For publishers, that relationship has value whether it is monetised through subscriptions, advertising or data. A small collection of articles serving an engaged and well-defined readership—on specialist software, for example—may be worth more than a large, flat archive in which much of the material has barely been read. The concentration and quality of the signal can matter more than the scale of the content collection.
Audience behavioural data
This creates an opportunity for the publishing industry, including organisations such as Ozone, to develop a different proposition. Open-web content can be scraped and resold by third parties under the contested protection of fair use, but the metadata associated with the audience for that content is a scarcer resource.
This matters as AI agents and large language models decide what to retrieve when a user submits a query. Simple keyword search helped power the first phase of consumer AI content ingestion, but there is now greater interest in high-quality, relevant and licensed material. Token costs and finite AI budgets increase the need to filter what enters an agent’s context window.
A publisher knows more than the content of a page. It knows who reads it, what they read next, whether they return and which subjects sustain their interest. Subject to appropriate privacy protections, that behavioural information can help explain whom a source serves and when it is likely to be useful.
Exposed as structured signals alongside the content, this information could become a form of first-party data for the agent era. Instead of offering only a page about living-room storage, a publisher could provide evidence that the source has been useful to a particular kind of reader with a particular need.
Grounding APIs, content marketplaces and access-control systems may provide the infrastructure through which this information travels. They are being built now, although the commercial arrangements, privacy protections and standards governing them are still uncertain.
Scraping provides a platform with content, but it does not provide the behavioural context surrounding it. That context exists in the relationship between the publisher and its readers. Combined with pressure on AI systems to prefer licensed, filtered and relevant sources, it may give publishers some leverage that they did not have in the previous platform cycle.
Nothing here is guaranteed. Audience metadata could become another opaque metric, or control could pass to a new group of intermediaries. Publishers will need to consider carefully what they expose, how it is measured and whether they retain control over its use.
They do not need to wait for these markets to mature before beginning the work. The foundations are familiar: understanding who engages with their content, why they return, which material earns their trust and what they do next. Decades of adtech and martech have already taught the industry how to analyse an audience. The next step is to apply that knowledge to agents as well as human readers.
Attention remains essential, but stripped of context it is just a commodity. Knowing the audience behind that attention gives us a more lasting signal of value.
