You Own the Output

The Fundamental Principle

This is the single most important idea in this module, so let us state it plainly: when you share AI-assisted work, you are responsible for its quality and accuracy. Full stop.

It does not matter that an AI model generated the first draft. It does not matter that the AI confidently stated a statistic that turned out to be fabricated. It does not matter that the tool is made by a prestigious technology company. The moment you put your name on something — the moment you send the email, submit the report, publish the article, or present the findings — it becomes yours. Your reputation. Your accountability.

This is not a burden unique to AI. It has always been the case that professionals are responsible for the work they deliver, regardless of which tools or assistants contributed along the way. A lawyer who relies on a junior associate's research is still responsible for the advice they give. An architect who uses computer-aided design software is still responsible for the structural integrity of their plans. AI does not change this principle. It simply adds a new tool to the mix — one that happens to be remarkably confident, occasionally wrong, and entirely indifferent to the consequences of its errors.

Verification Before Deployment

Before any AI-assisted work leaves your hands, it needs to pass through a verification process. The rigour of that process should match the stakes involved. Here is a framework to guide you:

Fact-checking. AI models can and do generate plausible-sounding information that is simply incorrect. They may cite sources that do not exist, misattribute quotes, confuse dates, or present outdated information as current. Every factual claim in AI-assisted work should be verified against a reliable, independent source. This is not optional. It is the bare minimum of responsible use.

Bias review. AI models reflect the biases present in their training data. This can manifest in subtle ways — the framing of an issue, the examples chosen, the assumptions embedded in the language, or the perspectives that are included and excluded. Read your AI-assisted work with a critical eye. Ask yourself: does this represent a balanced view? Are any groups or perspectives marginalised or stereotyped? Would this read differently if it had been written by someone with a different background or viewpoint?

Appropriateness for audience. AI does not know your specific audience the way you do. It cannot judge whether a particular tone is right for your team, whether a cultural reference will land, or whether the level of technical detail matches your reader's expertise. You need to make these judgements yourself, and adjust the output accordingly.

Logical coherence. AI can sometimes generate text that sounds polished but contains logical gaps, circular reasoning, or contradictions between sections. Read the work as a whole, not just paragraph by paragraph. Does the argument hold together? Do the conclusions follow from the evidence? Is there anything that would not survive a sharp question in a meeting?

The Journalist Test

Here is a useful mental exercise. Before sharing any significant piece of AI-assisted work, ask yourself: would this stand up to the same scrutiny as if I had written every word myself?

Imagine a journalist — or a demanding colleague, or an examiner, or a client — going through your work line by line. Could you defend every claim? Could you explain the reasoning behind every recommendation? Could you point to the sources that support every factual statement?

If the answer is yes, you are in good shape. If the answer is "mostly, but there are a few bits I'm not sure about," then you have more work to do before that output is ready to share.

When AI Gets It Wrong and Your Name Is On It

Let us be direct about the reputational reality. If you share AI-assisted work that contains errors, nobody is going to blame the AI. They are going to blame you.

Consider these scenarios:

• You send a client proposal that cites a competitor's revenue figure. The figure is wrong — the AI fabricated it. Your client checks, notices the error, and now questions the reliability of everything else in your proposal. The damage is not just to that document. It is to your credibility.

• You submit a report that includes a recommendation based on a mischaracterised regulation. The AI presented the regulation confidently but inaccurately. Your manager acts on the recommendation. When the error comes to light, it is your professional judgement that is called into question.

• You publish a blog post that includes a quote attributed to a well-known figure. The quote is fabricated — the person never said it. A reader notices and calls it out publicly. Your explanation that "the AI generated it" does not repair the reputational damage.

These are not hypothetical horror stories designed to frighten you. They are realistic scenarios that have already played out in various forms across industries. The common thread is simple: the person who shared the work bore the consequences, regardless of which tool produced the error.

Building a Personal Verification Checklist

Given everything above, it is worth developing a personal verification checklist that you apply consistently to AI-assisted work. Here is a starting point that you can adapt to your own context:

1. Have I read the entire output carefully? Not skimmed — read.
2. Have I verified every factual claim against a reliable source?
3. Have I checked that any cited sources actually exist and say what the output claims they say?
4. Have I reviewed the output for bias, stereotyping, or one-sided framing?
5. Have I ensured the tone and level of detail are appropriate for my specific audience?
6. Does the logical structure hold together from beginning to end?
7. Have I removed or rewritten any sections where I am not confident in the accuracy?
8. Am I comfortable putting my name on this and defending it if challenged?

If you can answer yes to all eight questions, your work is ready. If not, go back and address the gaps. The few extra minutes of verification are always worth it.