Reading tour operator and supplier contracts with AI: what it finds and what it misses

AI contract review is a language model reading a contract's text and returning locations, lists and comparisons. It finds clauses, quotes them with page numbers and compares two versions well. It does not judge legal effect, read annexes it was not given, or notice a clause that is missing.

The contract arrives as a 40-page PDF in English with three annexes and a separate price list, and the season starts in six weeks. A chat tool will summarize it in twenty seconds. The question is what you do with that summary, and what it quietly left out. After reading this guide you will know how to ask for findings instead of summaries, which clause families to ask for by name, how to compare two seasons of the same contract, and where the tool stops and your lawyer starts.

What is AI contract review good at, and what is it not?#

It is good at finding, listing and comparing. It is not good at judging. Treat the output as a map of the document, never as an opinion on it.

A language model reads the text you give it and writes text back. It holds no database of law, no view of your market and no memory of what a normal release period looks like in your destination this season. What it does have is patience: it will read all 40 pages every time, in any language, without getting tired on page 31.

The strong uses follow from that. Ask it to locate every place a clause family appears across the whole document, to list the dates, notice periods and percentages exactly as written, to translate a German or English clause into plain Turkish, or to set two versions of the same contract side by side. These are reading tasks, and reading is what the tool does.

The weak uses are everything that needs judgment. It cannot decide what a clause means for your hotel, whether it is enforceable, whether it is usual in the market, or whether the commercial trade-off is acceptable. It also cannot see what you did not give it: annexes, side letters, the operator's portal rules, or last season's emails agreeing an exception.

The real risk is tone. Fluent, confident prose reads like advice, and a confident sentence is not evidence. Ask for the source of every sentence the tool produces.

Why ask for the clause and the page instead of a summary?#

Because a summary compresses decisions into prose you cannot check, while a quoted clause can be verified in seconds. Ask for the article number, the verbatim text and the page, then open that page.

The prompt pattern is simple. Ask the tool to quote the clause word for word, give the article number and the page, and explicitly not to paraphrase. A model left to its own habits will summarize, because summarizing is the default behavior of a chat tool. The instruction to quote, with a location, changes the output from prose into evidence.

Then verify. Copy a distinctive phrase from the quote and search the PDF for it. If the phrase is not there, the quote was invented, and everything else in that answer is suspect. This check takes ten seconds and catches the single worst failure mode of these tools: text that looks like the contract but is not in it.

Two further questions surface what the tool cannot see. Ask which pages it did not read and which pages contained no text; this reveals scanned pages, image-only annexes and stamped amendments before you rely on a gap. And ask for the defined terms used inside the clause, because a release period read without the contract's definition of a season can mean the opposite of what a summary says.

Keep the quote, the article number and the page in your contracting file. A note you can defend two seasons later is worth more than a summary you cannot reconstruct. If the tool will only summarize, use it to decide which pages to read, never to decide what the contract says.

Any missing answer means the finding goes back to the pile.

What does AI miss in a tour operator contract?#

It misses what sits outside the text you gave it and what is not written at all: annexes, cross-references, definitions, scanned pages, and the clause that should be there and is not.

Annexes and price lists are usually separate files. Upload only the main body and the release grid, the rate table and the season calendar are invisible to the tool, and it will not always say so. A model answers from what it was given and rarely volunteers what it was not. Before anything else, list every file that makes up the agreement and decide which ones the question needs.

Cross-references carry meaning. A clause that says it applies subject to another clause cannot be read alone, and a summary that quotes only the first half inverts the rule. Ask the tool to resolve every cross-reference in the clause you care about and to list the ones it cannot resolve.

Definitions decide outcomes. Terms such as confirmed booking, room night or season may be defined in a schedule rather than where they are used, and the definition can change the arithmetic of a release period or a cancellation charge. Scanned pages, stamps and handwritten amendments have no text layer at all, so there is nothing to read; ask directly which pages came back empty.

The absent clause is the hardest miss. A model asked what a contract says about stop sale will answer fluently, but it will report that a contract says nothing about payment security only if you ask about payment security by name. Build the list of what should be there and ask about each item separately. Renumbering produces false differences too: a new version that moves clauses will look changed to a naive comparison even where the wording is identical. The comparison section below deals with that.

For a fuller picture of what these contracts contain beyond the room block, see our guide to the clauses beyond the allotment in tour operator contracts.

A clause-finding list for tour operator and allotment contracts#

Ask for these clause families by name, one at a time, and require a quote and a page for each. A single prompt asking for everything produces a summary; seven or eight narrow prompts produce findings.

Run the list in the same order every season, so this year's answers line up with last year's. Ask one clause family per prompt, because narrow questions get quoted text while broad questions get prose. For each family, ask three things: what the contract says, where it says it, and what the contract does not say. Record the answers with their pages in a single table you keep with the signed contract. Where a family is covered in an annex rather than the body, note which annex, because that is the file people forget to send to the lawyer.

If release periods are new to you, start with our guide to release periods and clauses to check in hotel allotment contracts.

How do I compare this season's contract with last season's?#

Ask for a clause-by-clause comparison rather than two summaries, and read the removals before the additions. A deleted guarantee is quieter than a new clause.

Label the files clearly in the prompt, for example A for last season and B for this season, and ask for three lists: added, removed, changed. Then ask for a table with four columns: clause, text in A, text in B, type of change. Prose comparisons hide the removals inside sentences; a table cannot.

Check every claimed change in both PDFs. Renumbering and reformatting create false positives, and a false positive costs you the trust you need for the real ones. If the tool flags a change, open both pages and read them. This is the same verification habit as the quote search, applied twice.

Ask a separate question about what stayed word for word but changed meaning because a definition, a currency or a season calendar moved elsewhere. Identical text with a moved definition is a real change that no word-level comparison will flag.

Treat a removal as a question for the operator, not a conclusion. Sometimes a clause was moved to an annex, sometimes it was dropped on purpose; the response tells you which. Keep the comparison table in the contracting file. It is the record of what you agreed to change and what you chose to accept.

Which contracts may you put into which tool?#

Only the ones the contract and your hotel's own policy allow, and only in a tool the hotel has approved for business information. A contract is business information before it is a text file.

Read the contract's own confidentiality clause first. It may restrict disclosure to third parties, and a cloud service is a third party. This is not a theoretical point in contracting: operator agreements often carry explicit confidentiality obligations, and the fact that a tool is convenient does not vary them. Confirm your own case with a legal adviser before uploading.

Follow the hotel's approved-tool policy: which tools are allowed, under which account, and who is permitted to upload. If no such policy exists, that is the first task, not the contract. Our checklist for choosing an AI chat tool for hotel staff covers what such a policy should say.

Prefer a business account where training on your content is switched off and retention is set. A free consumer chat window is the wrong place for a rate agreement. Remove what the task does not need: guest names, bank details, personal data, and any pricing you are not asking about. Where possible, extract the clause rather than uploading the whole file.

Keep a simple register: which document, which tool, which date, who uploaded it. It answers the question a supplier will eventually ask. Software and supplier contracts often name other companies and contain their information too, so the same rule applies with more care; see our guide to clauses to read in hotel software contracts before you sign.

Where does the lawyer take over?#

Wherever the question stops being what the text says and becomes what it means for you, whether it can be enforced, or what happens if it is broken.

Liability, indemnity, caps, termination for breach, penalties, governing law, jurisdiction and arbitration are legal-effect questions. A tool can quote them; it cannot tell you where you stand. The line is clean: the tool answers where the text is and what it says, counsel answers what it does.

The best use before counsel is preparation. Hand over the quoted clauses with their pages and the list of questions, so the lawyer's time goes on judgment rather than on reading 40 pages. This is where the clause-finding table earns its keep: a lawyer who receives eight quoted clauses with page references works on the substance immediately.

Ask counsel to confirm both the clauses the tool flagged as unusual and the clauses it reported as missing. A wrong absence is more expensive than a wrong summary, because an absence changes what you assume you are protected against. Tax and e-document treatment of invoices, commissions and currency belongs to your accountant, not to a chat window.

Nothing a tool produces is legal advice. Confirm your own case, the dates and any penalties with a lawyer or an accountant before you sign or rely on a clause.

What to do before the next contract arrives#

Take one contract you have already signed, run the clause-finding list against it, and see which families the tool cannot locate or quote with a page. That gap is your reading list for the next season, and it tells you exactly where a person still has to sit down with the PDF. Do this before the next contract arrives, not after.