How contract complexity can fool an AI tool, even when it sounds confident.

You've just received a job offer. The contract is 14 pages long. You paste it into ChatGPT and get back a clean, confident summary. Everything sounds balanced and reasonable. You sign. Three months in, you discover the restrictive covenant clause stops you working for any direct competitor within your sector for 12 months — and the termination clause you thought protected you actually works in reverse.

The AI wasn't lying to you. It was doing exactly what it's built to do: produce a helpful, fluent answer. The problem is that "helpful and fluent" isn't the same as "legally accurate".

AI gets contract review wrong in part because it predicts plausible-sounding responses, not legal effect. The most common failures: summarising clauses by their everyday meanings rather than the legal definitions in the document, missing risks that only appear when multiple clauses are read together, underweighting short clauses with major consequences, and being misled by the reassuring tone of professionally drafted documents.

Key Takeaways

• AI predicts plausible responses; it doesn't reason about legal effect

• Defined terms in a contract override everyday meanings; generic AI misses this

• Multi-clause risks (e.g. payment combined with termination) require cross-reading; AI reads linearly

• A confident-sounding summary is not the same as accurate legal analysis

• Use AI for extraction tasks (clause lists, specific terms); not for legal assessment

Who this is for

UK employees reviewing job offers, freelancers examining service agreements and NDAs, gig workers assessing platform contracts, and anyone who has been tempted to paste a contract into ChatGPT and trust the result.

Two questions come up most often: is it safe to upload your contract to AI tools, and does the AI actually give you accurate analysis? This article addresses the second question. For the first, see whether it's safe to upload your contract to ChatGPT.

How AI actually reads a contract

A lawyer slows down when language gets complicated. They notice when a reassuring clause has a quiet exception buried two paragraphs later. They track how a defined term in clause 2 rewrites the apparent meaning of clause 14. They read the whole document before forming a view on any individual clause.

AI models work differently. They predict the most useful-sounding response based on patterns in the text. That's a powerful capability for many tasks. But it means the model can be steered by the way a contract is written — not intentionally, not maliciously, but structurally.

In AI security, deliberate manipulation through embedded content is called prompt injection. You don't need the technical details. The practical lesson is simpler:

The way a contract is worded, structured, and framed can influence what an AI model pays attention to — and what it quietly glosses over.

Even without any malicious intent, structure and repetition can steer what the model weighs as important. A contract doesn't need to be designed to trick AI. It only needs to be complex enough that a general-purpose model smooths over the wrong detail.

Five ways AI misses the point in your contract

1. The summary sounds right, but the legal effect is wrong

Contracts often contain carve-outs: exceptions buried inside clauses that look protective on the surface. A clause that appears to limit your liability may contain a carve-out for "gross negligence or wilful misconduct" that effectively removes the protection in any situation where you're likely to need it.

A generic AI tool will typically summarise the headline protection. It may mention the carve-out in passing. But it's unlikely to flag that the exception significantly undermines the protection you thought you had.

2. Defined terms quietly rewrite everything

This is probably the most common source of AI error in contract review.

Legal contracts define their own vocabulary. Terms like "Gross Misconduct", "Confidential Information", "Services", or "Material Breach" look familiar but carry a specific contractual meaning set out in a definitions clause. A generic AI uses the everyday meaning of these terms. It doesn't track back to the definitions section and apply them consistently. The result: the summary sounds reasonable, but the legal effect is different from what the summary implies.

Example: Your contract says you can be dismissed for "Gross Misconduct" without notice pay. Sounds standard. Then you check the definitions clause and find it includes "any breach of company policy", defined so broadly that it covers things like using personal email on a work device.

3. The real risk is hidden across multiple clauses

Important legal risks often only become visible when you read two clauses together.

A payment clause that says invoices are due "within 30 days of written approval" combined with a termination clause that allows the client to terminate with 7 days' notice, for any reason, at any time: those two clauses together mean the client can terminate before approving your final invoice, leaving you with legitimate work unpaid and no contractual mechanism to force approval.

Neither clause, read in isolation, looks alarming. The problem is the interaction. Generic AI reads sequentially; it doesn't map interactions across sections.

Common combinations to watch:

• Payment terms + termination rights

• Notice period + garden leave provisions

• IP ownership + licence restrictions

• Confidentiality obligations + data processing language

• Liability cap + indemnity carve-outs

For a fuller list of the clauses most likely to cause problems when read together, see the risky clauses UK freelancers should always check.

4. Reassuring language distracts from the sharp edges

Professionally drafted contracts are designed to read well. The language is clear, the structure is logical, and the tone is neutral and professional. That presentation influences how AI tools respond.

If a contract contains 20 clauses that are reasonable and 2 that are one-sided, the overall professional tone will tend to pull the AI's assessment toward calm. The same effect happens with humans, which is why the best contract reviewers deliberately read contracts looking for what's missing, not what's there.

5. Short clauses with big consequences get overlooked

Non-compete clauses, IP ownership provisions, and automatic renewal terms are often a single paragraph. In a 14-page contract, a short clause doesn't command much attention from a sequential reader. But a non-compete that prevents you working in your industry for 12 months, or an IP clause that assigns everything you create to the client, are potentially more significant than any other clause in the document.

Common clauses AI tends to underweight: non-compete and restrictive covenants, IP ownership, termination for convenience, liability caps and carve-outs, governing law and jurisdiction, garden leave, payment trigger conditions.

For IP ownership specifically, see what UK freelancers need to know about IP in their contracts. For non-compete clauses, see whether non-competes in freelance contracts are actually enforceable.

What this means for your specific contract

The clauses where AI is most likely to give you an inaccurate picture:

Termination clauses. How much notice you get, what triggers termination, whether there's a payment in lieu of notice provision, and whether you retain the right to invoice for work already completed. These interact with payment terms in ways a summary won't capture.

Restrictive covenants. Non-compete clauses, non-solicitation clauses, and garden leave provisions. These are often short, often buried near the end of the contract, and often drafted more broadly than they're likely to be enforceable. A generic AI may flag them; it's unlikely to assess their enforceability accurately.

IP ownership. Who owns the intellectual property you create during the engagement, what happens to work in progress, and whether you retain the right to reference the work in your portfolio. In creative and technical freelance contracts, these clauses often assign more than the client needs.

Liability and indemnity. A liability cap that limits your exposure to the contract value sounds protective. An indemnity clause that requires you to cover the client's legal costs for any third-party claim related to your work can effectively remove that protection. Read them together, not separately.

A safer approach: how to use AI on your contract

AI tools are useful for contract review if you use them for the right tasks. Here's what they do well and how to get more out of them.

Ask for extraction, not assessment. Request a list of all clauses by type: "List every clause that restricts what I can do after the contract ends." This is a pattern-matching task that AI does well. Assessing whether those restrictions are enforceable is a different task, one that requires UK-specific legal knowledge.

Ask for direct quotes. "Quote the exact wording of the termination clause, including all sub-provisions" gives you the actual language to review yourself. A paraphrase loses the legal precision.

Test the AI before you trust it. Ask the AI a question you already know the answer to from the contract — the notice period, the start date, or the daily rate. Ask it to quote the relevant clause number and exact wording. If it gets that wrong, hedges, or paraphrases instead of quoting, treat the rest of the analysis with significant caution. This is a quick way to calibrate how well the model is actually reading your specific document.

Manually verify high-impact sections. For any employment or service contract, always read these yourself regardless of what the AI says: termination, liability caps, indemnity, IP ownership, restrictive covenants, renewal provisions, payment triggers, and the definitions section.

Check every capitalised term. Any term that's capitalised in the contract body (Confidential Information, Gross Misconduct, Deliverables) has a specific definition somewhere in the document. Find it and read it before you trust any summary of the clause that uses it.

Why purpose-built contract review is different

Tools built specifically for legal documents differ from generic chatbots in ways that matter.

Clause-by-clause analysis means every section gets individual attention, not just the ones that appear most prominently in a summary. Jurisdiction-specific insights mean the analysis is grounded in UK contract law, not general legal knowledge from multiple jurisdictions. Cross-clause risk identification means the tool can flag when a payment clause and a termination clause interact in a way that disadvantages you. And plain-English explanations mean you understand what each clause actually does, not just what it says.

Privacy is also different. With a general-purpose AI tool, you need to anonymise your contract before uploading to avoid data and confidentiality risks. For a guide to doing this safely, see how to anonymise your contract before uploading to any AI tool.

The bottom line

The question to ask isn't just "Is it safe to share my contract with this AI?"

It's also: "Is this AI actually reading my contract in a way that preserves the legal risk?"

That's a much harder bar to clear. It's why AI contract review needs more than a general-purpose chatbot.

Ookulli reviews your contract from £10, clause by clause, UK-specific, with your data kept private from the start.

Frequently asked questions

Can ChatGPT review a contract?

It can summarise and extract information, but it shouldn't replace specialist review for any contract you're about to sign. Generic AI may overlook defined terms, carve-outs, cross-references, and risks that are only visible when you read multiple clauses together. It's most useful for extraction tasks; least reliable for legal assessment.

Can I trust ChatGPT to review my employment contract?

As a starting point for orientation, it can help. For a signing decision, no. Employment contracts contain termination provisions, restrictive covenants, IP clauses, and jurisdiction-specific protections that require UK-specific analysis. A generic tool won't reliably flag problems with these, and it may actively reassure you when the contract warrants concern.

Why does AI sound so confident even when it gets things wrong?

AI models are trained to produce helpful, clear, fluent responses. Confidence in tone reflects the model's output style, not the accuracy of the legal analysis. A response that sounds authoritative about a contractual risk may still be missing the defined term that changes what that clause actually means.

What is prompt injection and does it affect contract review?

Prompt injection is when content influences an AI model in ways that change what it prioritises or how it responds. In contract review, the practical lesson is that wording, structure, and framing can steer a general-purpose model away from the clauses that carry the most legal risk — even when no one is trying to trick it.

What contracts is AI most likely to miss important clauses in?

Longer contracts with complex defined terms, contracts that mix confidentiality with restrictive covenants (a common structure in client service agreements and NDAs), and contracts with liability caps that interact with broad indemnity provisions. If your contract has any of these, treat any AI summary as a starting point only.

Does the way a contract is written affect what AI picks up?

Yes. Professionally drafted contracts read well: clear language, logical structure, neutral tone. That presentation can pull AI analysis toward calm even when individual clauses are one-sided. Contracts that use a lot of capitalised defined terms are also more likely to trip up generic AI, because the model uses everyday meanings rather than the specific legal definitions in the document.

Is AI contract review legal in the UK?

Yes, using AI to assist with contract review is legal. There's no restriction on using AI tools for personal or commercial document analysis. The questions are practical, not legal: whether using a public AI tool breaches the confidentiality clause in your contract, and whether uploading a document containing personal data is compliant with UK GDPR. A tool built for legal documents sidesteps both concerns.

What should I do instead of asking AI for a summary?

Ask for extraction, not assessment. Request a list of all clauses by type, ask for exact quotes from high-risk sections, and verify those sections manually. For assessment of enforceability, especially for restrictive covenants, IP clauses, and liability provisions, use a tool built for legal review or seek brief legal advice.

Review your contract with Ookulli from £10 — clause by clause, UK-specific, and built for legal documents.

This article is for informational purposes only and does not constitute legal advice. If you have specific concerns about your employment contract or legal rights, you should seek advice from a qualified solicitor.