AI & Technology
What Does an AI Watermark Actually Tell You?
An AI watermark can tell you who made it. It cannot tell you whether it is good.
By Fred Soller · 2026-09-14
Knowing where AI-generated content came from matters. It still does not tell us whether work is accurate, useful, or worth trusting.
The technology keeps moving
Anthropic has released Claude Sonnet 5 and Claude Opus 5. OpenAI continues to improve GPT-5.6 and is previewing a much faster version of the service. The names and technical details will keep changing, but the direction is easy to see: these tools are getting better, faster, and easier to put to work.
At the same time, both companies are trying to make AI-generated content easier to identify.
I am not an expert on the technology behind this, but the direction is worth paying attention to. Anthropic recently announced that it will begin placing invisible markers in Claude-generated text. OpenAI is doing something similar with images and is expanding those efforts into audio and video. The goal is straightforward: help people recognize content created by AI, even when there is no visible label.
That is useful, but only answers one question: Was this made by AI?
It does not tell us whether the content is accurate, useful, or aligned with the business. That is the part leaders cannot afford to miss.
Knowing the source is not proof
A watermark is a receipt, not a seal of approval.
If AI produces a bad market analysis, a weak sales proposal, or a confidently wrong forecast, knowing which model created it does not make the work any better. It only makes the source easier to trace.
“Garbage in, garbage out.” Feed AI incomplete customer data and it can build a polished but unreliable customer profile. Ask it to forecast revenue using inconsistent sales stages and close dates, and it can produce a very professional-looking answer that is still wrong. Give it a vague value proposition and it will create more vague messaging, only faster.
The output can look impressive and still be garbage.
Better models raise the stakes
These products are no longer just writing assistants. They can research accounts, analyze large amounts of information, create proposals, update systems, and carry out work across several steps.
When AI was mainly drafting emails or summarizing meetings, a bad answer was usually an inconvenience. When it starts prioritizing leads, recommending pricing, changing CRM records, or contributing to a forecast, a bad answer becomes a business problem.
The better these tools get, the more attention leaders need to pay to the process behind them. AI does not clean up a broken process simply because it runs the process faster.
The real risk is bad work that looks credible
The biggest risk may not be an obvious fake. It may be a polished piece of work that moves through the company without anyone questioning the assumptions behind it.
Imagine an AI-generated territory plan built from stale account data. The presentation looks sharp. The recommendations sound confident. The source is clearly identified.
The sales team can still spend the quarter chasing the wrong accounts.
Or imagine a pipeline forecast built on poorly defined stages and unreliable close dates. AI can calculate probabilities and write a strong executive summary. It cannot create the sales discipline that leadership never put in place.
The board can still receive a misleading forecast.
That is what makes it more dangerous today. The garbage no longer looks like garbage. It arrives formatted, summarized, and ready for the executive meeting.
Four questions leaders should ask
Companies should not slow down their use of AI. They should become more disciplined about where and how they use it.
Before putting AI into an important workflow, leaders should ask:
Is the underlying data trustworthy?
AI cannot create a reliable view of the business from duplicate accounts, missing customer history, or inconsistent definitions.
Is the process clear enough to automate?
If the team cannot agree on how to qualify an opportunity or build a forecast, AI will not settle the debate. It will simply run one version of a process no one agreed to.
Who owns the outcome?
Someone still needs to own the accuracy and judgment behind the final work, especially when it affects customers, employees, capital, or reputation.
Are we measuring results or activities?
The number of prompts, licenses, and automations tells us very little. Better conversion, shorter sales cycles, improved forecast accuracy, stronger retention, and lower cost to serve are what matter.
AI will scale the system you already have
OpenAI and Anthropic are building more powerful models and better ways to identify what those models create. Both efforts are worthwhile.
Neither changes the basic leadership challenge.
AI will not fix an unclear strategy, a poorly defined ideal customer profile, unreliable data, or inconsistent execution. It will help a disciplined company move faster. It will also help a confused company create more confusion at greater speed.
A watermark may tell us that AI created something. It cannot tell us whether the organization using it knows what it is doing.
Before asking how quickly AI can scale your go-to-market engine, ask a harder question: Is the engine worth scaling?
Sources
Fred Soller is a Co-Founder & Managing Partner at Big Wheel Performance. He helps enterprise and mid-market companies build and scale global revenue organizations by driving GTM alignment, execution discipline, and exit readiness.