, Nils-Erik Jansson
Reviewing an AI answer in ten minutes: five steps
An AI answer is reviewed in five steps: the sources, case law and statutory provisions, figures and dates, what is missing, and the tone. Responsibility for the advice is the same whether or not the answer came from a tool, and an answer is hard to review when it is well written, confident and arrives in a few seconds. This guide sets out the five steps and the errors that look right and therefore slip through.
Updated by Nils-Erik Jansson
Practice piece
Invented for this exercise
The question put to the tool
A colleague pasted an HR matter into a personal AI account. What do we have to do?
The answer. Mark every sentence you would check before using it.
Five errors are planted, one for each step. The answer is not legal advice. Nothing is sent or stored.
Errors that look right
01
The case with a plausible name and number that does not exist.
02
The section that exists but whose wording has changed.
03
The foreign rule that has crept into an answer on Swedish law.
04
The deadline that applied last year.
05
The summary that is correct sentence by sentence and lacks the decisive condition.
06
The amount that is right in the wrong currency.
07
The confident answer to a question where the law is unsettled, with no mention of the uncertainty.
Background
Why a method
The guidance of the Swedish Bar Association (Sveriges advokatsamfund) says that output must be reviewed critically, in particular sources, case law and statutory provisions, and that the lawyer needs to know how the tools work and where they fall short [1]. That applies to every lawyer who stands behind a piece of advice. Most are willing. The problem is that an AI answer is written to be read rather than checked, and a review without a method becomes a read-through that approves whatever sounds right.
Five steps, in order. The last two are what separate a lawyer from a proofreader.
| Step | What is reviewed | The question |
|---|---|---|
| 1. The sources | Cited sources | Does the source exist, and does it say what the answer claims? |
| 2. Case law and statutes | Cases, sections, wording | Does it exist, at the right court level and in the right wording, and is it about what the answer says? |
| 3. Figures and dates | Amounts, percentages, deadlines, entry into force | Does each figure match the source? |
| 4. What is missing | Exceptions, transitional rules, conditions | What has the answer left out that would change the conclusion? |
| 5. Tone and responsibility | Certainty and judgment | Is the answer as certain as you can be, and is it your judgment? |
Step 1: the sources
Does the source exist, and does it say what the answer claims? Open every source cited. If it does not exist, strike the statement. If it exists but says something else, strike the statement and note that the tool has a tendency on that question. If it exists and says what is claimed, mark the statement as checked.
If the answer has no sources at all, ask for them before doing anything else. An answer without sources cannot be reviewed, only believed.
Step 2: case law and statutes
Case law: does the case exist, at the right court level, with the right year, and is it about what the answer says? Invented cases often have plausible names, plausible numbers and a summary that fits the question perfectly. The perfect fit is what should make you suspicious. Look the case up in the legal database you use, not in the tool.
Statutes: the right section, the right subsection, the wording in force? Common errors are the right section but the wrong subsection, wording that has been repealed or amended, and Swedish law mixed with foreign law where the question concerns an area where both are online, such as contract law and data protection. Check the wording in the official source, with its date.
Step 3: figures and dates
Amounts, percentages, deadlines, dates of entry into force. Check every figure against the source. The tools can state deadlines and commencement dates with confidence even when a date has moved. The application dates of the EU AI Act are an example of dates to check against the source.
Step 4: what is missing
Now read the answer as the other side would. Which exception is not mentioned? Which transitional rule? Which condition in the contract would change the conclusion? Which fact in your matter did the tool not know about, because you did not write it? That what is there is correct says nothing about what is missing. This is the hardest step, and it requires the reviewer to know the area.
This is also where you catch the polished summaries that leave out a condition. The summary is right in every sentence and wrong as a whole, because the condition that decided the matter dropped out. Read the original on the points that carry the conclusion.
Step 5: tone and responsibility
Read the answer as if you had written it and had to stand behind it. Is it as certain as you can be? The tools can phrase things more confidently than the sources allow. Is it your judgment or the tool's? Rewrite whatever is judgment in your own voice and with your own reservations. And ask yourself whether the recipient would notice that it was written by AI. If so, it is not finished.
Errors that look right
The case with a plausible name and number that does not exist. The section that exists but whose wording has changed. The foreign rule that has crept into an answer on Swedish law. The deadline that applied last year. The summary that is correct sentence by sentence and lacks the decisive condition. The amount that is right in the wrong currency. And the confident answer to a question where the law is unsettled, with no mention of the uncertainty.
When ten minutes is not enough
When the answer will be the basis for advice in a dispute, a major contract or an employment matter, ten minutes is a first sift, not the review. Then step four is done properly, with your own research, and the tool's role is that of a first drafter, not a source. These are also the matters where the question of what may be entered at all comes first; see the checklist.
How HOVR Legal works with this
The review method is part of the training in the setup of Claude for Legal and of the AI literacy training, and HOVR Legal writes the review rule into the guidelines; see the guide on an AI policy for the legal department. Tools that show a source for each statement make step one possible, and that is one of the requirements HOVR Legal tests in the demo on your own contracts.
Frequently asked questions
- Does every answer have to be reviewed, even a summary for my own use?
- Anything that leaves the department or forms the basis of advice is reviewed. A summary you read yourself to get your bearings does not need the full review, but it must not be passed on until steps one and two have been done. Write the rule into the policy like this: no AI-generated content goes to anyone else unless the sender has read it and checked the sources.
- Can the tool review its own answer?
- In part. Asking the tool to give a source for each statement, or running the same question again in a new chat and comparing, catches some errors. But the tool does not invent fewer sources because it is asked to check them, and two wrong answers that agree are still wrong. The check against the original source is done by a person.
- How do we write the review into the policy?
- As a three-line rule: what is checked (sources, case law, statutory provisions, figures), by whom (the person using the answer), and before what (before it leaves the department or forms the basis of advice). Add spot checks; one suggestion is that ten reviewed answers per quarter are looked at afterwards. See the guide on an AI policy for the legal department.
Sources
- [1]Vägledning om användning av generativ AI i advokatverksamhet, Sveriges advokatsamfund, 12 juni 2025, in Swedish; Swedish Bar Association guidance on generative AI, section 3 on responsibility for advice and review of output
Insights on the topic
Back to AI for legal
More under AI for legal
- Data protection package for AI tools in legal work
Fixed-scope engagement: transfer impact assessment, impact assessment where needed, draft addendum, policy and training. For the AI tool you use or consider.
- AI operating model for the legal department
Fixed-scope engagement for legal departments. You get an operating model for how AI is used, governed and followed up, whatever the tool.
- AI literacy training for lawyers: two hours
Two hours of training for your legal department or firm on your own tools and documents, documented as a measure under Article 4 of the AI Act.
- Data protection checklist for generative AI in legal work
A checklist for law firms and legal departments: five questions to answer before the first client data goes into an AI tool, and what counts as a pass.
- AI policy for the legal department: what it should contain
Outline of an AI policy for a legal department or law firm: permitted tools, what may be entered, review, responsibility, excluded matters and training.
- AI policy template for the legal department
An editable AI policy template for a legal department or law firm in nine sections, with comments on what to adapt and why.
- The EU AI Act for in-house counsel: what applies to users
What the EU AI Act requires of a legal department that uses generative AI: AI literacy since 2025, transparency since 2026 and when the high-risk rules apply.
- Transfer impact assessment for AI tools, step by step
The six steps of a transfer impact assessment for a US AI tool: map, choose a transfer tool, assess the country, supplement, document, review.
- DPIA for generative AI: when it is required and how
When a legal department or law firm must carry out a DPIA under Article 35 GDPR for an AI tool, using the Swedish regulator IMY's nine criteria.
- Personal data in prompts: legal basis and information
When a lawyer writes about an employee, a counterparty or a customer in an AI tool, personal data is processed.
- AI incident response for legal teams
What a legal department does within 72 hours when a member of staff has entered data into an AI tool that should not have gone in.
- Anonymise before prompting: what works
How a lawyer removes personal data before text goes into an AI tool, and the difference between anonymisation and pseudonymisation.
- Swedish MBL and AI tools: must you negotiate first?
Swedish law: when an AI tool triggers the duty to negotiate under section 11 MBL, when informing under section 19 is enough, and with no collective agreement.
- Shadow AI in the legal department: one week of work
Staff use ChatGPT and Claude on private accounts for work. The legal department's week: map it, stop the worst, offer an approved tool, write the rule.
- Legal privilege and AI: the Swedish Bar guidance
What the Swedish Bar Association's guidance on generative AI requires for confidentiality: tools that do not train on client data, and terms read in full.
- Contracts with AI vendors: asking beyond standard terms
What a legal department or law firm asks of an AI vendor beyond the standard terms: authority requests, sub-processors, deletion, anonymisation, exit.
- On-premise or cloud AI for legal work
When a legal department or law firm should run AI locally and when the cloud is right: what local operation solves, what it costs, what it leaves open.
- AI literacy training for the legal department
What a two-hour training for lawyers using generative AI covers: what the tool does and does not do, your rules, practice on your own documents.
- AI guidance for lawyers in the EU and the UK
What the CCBE, the SRA, the Law Society, the Bar Council and the Swedish Bar Association require of lawyers using generative AI, and where they differ.
- The general counsel's AI report to the board: two pages
What the board needs to know about the company's AI use once a year, and what the general counsel can answer for.