In conference breakout rooms, industry chats, or in hushed exchanges between research leads, there is a conversation happening across the knowledge and information profession right now – and it goes something like this:
“Our staff are already using AI tools to enhance our research services, and we want to expand that further. But when we go to our data vendors and ask how we can do this properly, using the licensed content we pay significant sums for, we hit a wall.”
That wall has a name. I’m calling it “Vendor AI Stalemate,” and it is threatening the relevance of both the knowledge and information profession, as well as the data vendors we depend on.
The Vendor’s Problem Is Real. But So Is the Cost of Inaction.
To be fair to vendors, their hesitation is not without reason. The commercial models that have underpinned the information industry for decades, such as per-seat licensing and usage-based pricing, were designed for a world where a human opened a database, ran a search, and read an article. They’re not designed for a world where an AI model ingests thousands of documents simultaneously, synthesises them into a single output, and delivers that output to a user who may never see or even cite the original source.
Vendors are right to ask questions such as: How is their data being used? How many times is it effectively “consumed” in a single AI query? What happens to their content once it is embedded in a model’s outputs? How do they charge for something they can no longer see? How do they write contract terms that protect intellectual property when the nature of AI usage makes traditional usage metrics meaningless?
These are legitimate concerns – building a sustainable chargeback model in this environment is a complex challenge. Vendors are currently spending significant time in internal working groups, legal reviews, and commercial strategy sessions trying to work out what AI means for their business model.
While Vendors Deliberate, Users Are Moving On
However, the need data for data does not pause while vendors figure out their pricing model. Users are simply going to find it somewhere else. Everyday across organisations of every size and sector, employees are turning to AI tools and asking the same questions they once directed to research teams or licensed databases. The results are imperfect. The sourcing is opaque. The data has often not been validated or verified. But it is fast, it is free, and it is “good enough.” “Good enough,” possibly the two most dangerous words in the industry and something that has always kept research leaders awake at night, is even more front of mind right now.
“Good enough” is not a quality threshold. It means: “I know this might not be perfect, but I cannot access what I need through legitimate channels quickly enough, so this will do.” It means that the value of premium, licensed, curated information, the very thing vendors have built their businesses on, is being eroded not by a competitor offering something better, but by a free alternative that is providing enablement.
A vendor that spends too long calculating exactly how much their data is worth may discover that the market has already moved on.
The Research Professional Caught in the Middle
If this dynamic is frustrating for vendors, it is untenable for those of us working in research services. Our roles have always been grounded in a commitment to data quality. We are the people who care about where information comes from. We know which sources are credible and which are not. We know which databases have rigorous editorial standards and which aggregate content indiscriminately. We understand the difference between a primary source and a synthesis, between a verified statistic and an extrapolation. We are, in the truest sense, ambassadors for data integrity.
That role matters more now than ever. In a world flooded with AI-generated content, where hallucinations are a known risk and provenance is increasingly difficult to verify, having people in organisations who champion the reliability of information is a necessity.
In practice, when working with vendors who have not yet found a workable model for AI integration, we cannot tell our staff that the data being used within AI tools has been validated, verified, or sourced from a credible publication. But equally, we cannot point them to an approved pathway for AI-assisted research that gives us confidence in the output quality. So instead, we become something we never wanted to be: “The team that says no”. No, you cannot use that tool in that way; no, you cannot use that licenced data in your AI agent; no, we have not agreed terms with that vendor.
And when you are the team that always says no, users stop asking. They go around you. They self-serve. They use the free tool, get a good enough answer, and move on. And slowly, imperceptibly, the research function stops being seen as an enabler and starts being seen as an obstacle.
That is an existential risk, for individual teams and the profession as a whole.
What a Potential Model Looks Like
While the challenge may feel too large, too costly, and too legally complex to resolve, some vendors are showing that progress is possible.
Imperfect terms agreed today are better than no terms at all. The ideal commercial framework for AI use of licensed content may still be some way off, but an interim model that recognises AI use is already happening, seeks to define it as clearly as is currently feasible, and establishes a basis for fair compensation is far preferable to a vacuum.
Transparency must also become part of the operating model. An understandable concern that most vendors share is the lack of visibility into how their data is being used. It is technically possible to develop AI-native access models in which usage is logged, queries are attributed, and outputs can be traced back to source material. Some vendors are already beginning to explore them, and they will be the vendors better placed to develop sustainable pricing models. Crucially, such models will also give research professionals the ability to say with confidence where the information in an AI-generated output came from. However, as it stands, of course, there will always be a risk that data is used far outside the bounds of a contract, but that risk is not new. Knowledge and information professionals have long played a critical role in promoting, educating, and reinforcing proper use of licensed content within their organisations. They will continue to do so. If vendors are willing to trust that partnership, even in an imperfect environment, it creates the conditions for progress.
Our information and knowledge professionals must speak with greater unity, projecting a consistent message: that we want to use licensed data legitimately in AI contexts, that we do understand the complexity, and that we are committed to making this work. There will be bumps in the road. But working together as an industry with the clear intention to do right by each other, to review and refine as we progress, and a spirit of collaboration with our valued vendors, should be the overwhelming call to action.
Much of the current deadlock is framed around how much vendors should be paid for AI use of their content, with some vendors putting forward +100% increases because of the potential for increased data usage. But the value of data does not rise dramatically simply because consumption increases. The more important question is: what value does verified, licensed, high-quality data bring to AI outputs compared with unverified free alternatives? The final deliverable – which is the insight itself – has not fundamentally changed. What has changed is the route taken to produce it. That is why vendors need to be willing to take measured risks with customers, test new approaches, and pilot workable models before the market moves on without them. If there is a demonstrable example that AI has enabled firms to derive significantly more value because they can do something different with the data and get a new type of insight, then that can be used as levers for future pricing. Until then, vendors must be realistic about AI-related data usage.
For many vendors, the usual delivery model is offering AI-powered features within their own platforms, often as premium add-ons. Some vendors therefore argue that subscription prices should rise because AI tooling has reduced the time needed to undertake research. However, data analysis is not always about the time taken. The client may in fact be utilising their productivity gains to improve deliverables, redirecting the time not recouping it.
And whilst vendors may be looking to benefit commercially from locking AI utility behind their own walls, their customers have invested in building integrated, multi-source AI capabilities to unlock productivity gains. If vendors want to persist with pricing increases based on efficiency, they need to get their data out of the black box. The biggest value they offer to clients is the data itself, not their platform or the productivity gains coming from the client’s AI capabilities and infrastructure.
The Window Is Not Open Indefinitely
I want to be clear about the stakes here, because I think they are sometimes underestimated by both sides.
For vendors: The window in which premium licensed data is seen as meaningfully superior to free alternatives is real, but it is not permanent. User habits are forming right now. Workflows are being built. Organisations are making decisions about which tools to embed and which to abandon. Every month that passes without a workable AI integration model is a month in which those habits calcify around tools that do not include a vendor’s licensed content. Recovering market position later is significantly harder than retaining it now.
For the information and knowledge professionals: Our value proposition has always rested on our ability to access, assess, and deliver information that others cannot easily find or evaluate. If the AI tools our organisations use are locked out of the best data sources because the commercial framework does not exist, we lose a critical part of that proposition. And if we respond to that by simply becoming the gatekeepers withholding data, we accelerate our own marginalisation.
The resolution of this stalemate is not just a commercial negotiation. It is a question of whether the information profession, vendors and practitioners alike, can emerge from the AI transition as a central and credible part of how organisations access knowledge, or whether it gets bypassed entirely by tools that are faster, freer, and “good enough”.
I know which outcome I am working towards.
—
The author leads a research function and will be speaking at the KIMRA Conference on 3 June on the topic of vendor relationships, AI integration, and the future of licensed data in research services.
