An AI-generated comparison table is ready to publish when each row asks the same question about every option, each answer can be traced to a source, and missing information is labeled for what it is. Start with the claims in the cells. Clean headings and tidy checkmarks cannot fix a price with the wrong time basis or an invented “No.”
Write the question behind each row
A row headed “Cost” could mean the hourly price, the minimum amount due, or the total for the visit a reader is planning. Decide which question the row answers before editing any cell. If two questions matter, give each its own row or state the comparison scenario above the table.
A useful row definition records the attribute, the unit or allowed wording, the scope, and the date or version of the sources. For example: “Room charge in dollars for a two-hour Saturday booking, before any charges the booking sheet does not specify.” If you cannot describe what makes two answers comparable, the reader cannot safely compare them either.
Keep the options at the same level
A room, a building, and a bundled event package might each be useful options, but they cannot all be treated as identical room rentals. Likewise, a price for one date does not establish a price for every date. Use the name and version from each source, and explain any boundary that affects more than one cell.
Keep a source ledger beside the draft
For every consequential cell, write down the row and option, the exact source location, the relevant wording or value, and the editorial decision. Use a line or section reference that another person can open. When the table has been revised twice, the ledger should still tell you whether a number was copied, calculated, or left unresolved.
Use distinct labels for distinct states:
- Confirmed: a source supports the statement as written.
- Conditional: the source supports it only with a requirement or exception.
- Not stated: the source you checked does not answer the question.
- Not available: a source explicitly says the option does not provide it.
- Not applicable: the criterion does not apply to this option.
The UK Analysis Function's guidance on missing cells explains why one shorthand can be ambiguous when several states are possible. For a short public table, spell out the status rather than giving readers a code to look up.
Keep derived values reproducible
When you convert a quoted rate into a comparable total, record the original rate and calculation. Check minimums and what is included first. A mathematical conversion is valid only under the conditions it uses. For more involved claims, use the guide to checking numerical comparisons in AI writing.
Work through a fictional room comparison
These invented booking notes are a teaching fixture, not real venue prices or a test of a model.
- Oak room, booking sheet §2: $30 per hour, two-hour minimum. Projector included.
- Maple room, booking sheet §3: $55 for a two-hour block. Projector available by advance request. The sheet does not mention an equipment fee.
- Cedar room, booking sheet §1: Two-hour block, price on request. No projector.
Suppose the draft shows prices of $30, $55, and $0, and a projector row of Yes, Yes, and No. Only the last projector cell is fully supported as written. The Oak rate is not the two-hour charge. The Maple projector is conditional. Cedar's unknown price is not zero.
A corrected grid might read:
| Attribute for a two-hour booking | Oak | Maple | Cedar |
|---|---|---|---|
| Room charge | $60 ($30 × 2) | $55 | Price on request |
| Projector | Included | Advance request required | Not available |
| Separate projector charge | None; included | Not stated | Not applicable |
The cell notes explain why the room charge is comparable for the first two options, while the total cost of a projector booking still cannot be compared. The room comparison can be useful without declaring a winner. A reader who needs a projector has a concrete question to ask Maple before booking.
Preserve exceptions where readers will look
“Available” is too broad if advance notice is required. Put the condition in the cell or immediately beside the claim. A general note that “terms may vary” does not tell a reader which option has the exception. If a condition is too long for a cell, link to the relevant details and leave a short summary in the cell itself.
Edit for scanning after the claims are checked
Use short, descriptive headings and parallel entries. Put a shared unit or scenario in the heading or introduction so the reader can interpret any cell without guessing. The Australian Style Manual's table guidance covers units, captions, and source notes; Microsoft's guidance on tables recommends parallel entries and brief cells for responsive layouts.
Open the rendered page and try reading across one row on a narrow screen. The option names and attribute labels should remain understandable. Use table headers rather than an image of a grid, and keep any yes/no or risk distinction readable without color. If the information is really just a list of three unrelated descriptions, use a list instead.
Check what the prose claims about the table
Read any sentence above or below the grid that says “cheapest,” “includes,” “all,” or “only.” In the example, Maple has the lower stated room charge among the two known charges. “Maple is cheapest” goes beyond the evidence because Cedar's price is unknown and Maple's equipment charge is unconfirmed. If the document needs a recommendation, handle it separately with the guide to reviewing an AI-generated decision brief.
A final review question is whether someone could make the wrong choice because a label, time basis, or exception got lost. If so, repair the table where they would make the comparison, not just in the background notes.
FAQs
Should I use a dash for an unknown value?
Use words when possible. “Not stated” and “not applicable” answer different questions. Zero is a value and needs evidence of its own.
Can I ask AI to fill in the missing cells?
Ask it to identify the gaps or point to possible source passages. Verify those passages yourself. Keep an unsupported cell labeled as unknown until a source or responsible owner settles it.
About Humanizer.help Editorial Team
The Humanizer.help editorial team writes practical AI writing guides for students, ESL writers, and academic drafts that need clearer, more natural English.
Related Articles

How to Check Percentages and Numerical Comparisons in AI Writing
Check denominators, percentage points, relative changes, and rounding with worked examples before accepting numerical claims in an AI-assisted draft.

How to Review an AI-Generated Decision Brief Before Leadership Review
Check an AI-generated decision brief for a real choice, comparable options, evidence, tradeoffs and unresolved assumptions before sending it.
