A comparison table looks clear on your monitor. Then you need a plain-text email, a spoken briefing, or a short paragraph for a page that cannot accommodate the grid. You ask AI to summarise it and get a confident sentence that merges two columns, drops a qualification, and declares a winner nobody asked for.
The writing problem is preserving relationships when the layout disappears. This guide complements our same-scale comparison guide: first establish that the comparison is meaningful, then translate it into prose a reader can understand without seeing the table. The venues and figures below are fictional, supplied solely for this example.
Start with the table's question
Suppose a community group is choosing between two rooms for a workshop with 18 attendees. Approved notes say Cedar Room has 24 seated places, costs 120 credits for a three-hour booking, and includes a projector. Willow Room has 20 seated places, costs 90 credits for a three-hour booking, and offers a projector for another 15 credits. Both prices cover the same booking period. Step-free access has been confirmed for Cedar; Willow's access information has not yet been confirmed.
The reader needs capacity, comparable cost, equipment, and the unresolved access detail. A paragraph about the venues' atmosphere would be an invention because the notes contain no description of it. A claim that either room is available next Saturday would also exceed the source. Define the decision before asking the model what to keep.
Make hidden labels explicit
Tables let a short cell such as “120” borrow meaning from a column heading. Prose has to supply that context directly. “Cedar costs 120 credits for three hours” preserves the currency unit and booking duration. “Cedar costs 120” makes the reader reconstruct both.
The same applies to missing information. An empty cell or dash might mean unavailable, unknown, not applicable, or simply not entered. Resolve its meaning with the source owner. In this example, Willow's access status is explicitly unconfirmed. Writing “Willow has no step-free access” would turn missing evidence into a negative fact.
Before drafting, make a short list of the labels that every relevant value needs: room name, seated capacity, booking duration, currency unit, projector condition, and access status. This is a preparation aid, not extra text that must appear mechanically in every sentence.
Compare a weak summary with a usable one
Weak draft: “Willow is the best choice: it is cheaper, seats everyone comfortably, and has the same equipment as Cedar. Cedar is the accessible premium option.”
The source establishes neither comfort nor an overall winner. Equipment is offered on different price terms. Calling Cedar “premium” invents a market position, while the access wording implies a settled contrast that the notes do not support.
Revised draft: “Both rooms have enough seated places for 18 attendees: Cedar has 24 and Willow has 20. For a three-hour booking with a projector, Cedar costs 120 credits and Willow costs 105 credits, including its 15-credit projector charge. Cedar's step-free access is confirmed; Willow's access information still needs confirmation before a booking decision.”
The revision states the comparison basis and keeps the unresolved question visible. The 105-credit total is an editorial calculation from 90 plus 15; check that arithmetic yourself. Do not rely on the model's confidence or silently add charges the notes do not mention.
Choose the order around the reader's task
If confirmed step-free access is a requirement, lead with that point rather than putting the lower price first. For a neutral briefing, present capacity, total cost on the same basis, and the unresolved detail. Either order can be accurate, but they direct attention differently.
A useful opening could be: “Cedar meets the confirmed access requirement. Willow's access needs checking; its lower projector-inclusive price does not resolve that question.” This is conditional on the group's stated requirement. Without that requirement, do not invent one just to make a stronger recommendation.
If several options and conditions make the paragraph hard to follow, use one labelled paragraph per option or a short list. Compressing every fact into one elegant sentence can make the relationships harder to recover.
Ask for a draft and a separate uncertainty list
Try this prompt: “Convert these approved room notes into a plain-text comparison for a workshop with 18 attendees. Preserve room names, units, booking duration, equipment charges, and unconfirmed information. Compare projector-inclusive totals and show any calculation separately. Do not infer comfort, availability, accessibility, or an overall winner. Put unresolved questions outside the publishable paragraph.”
If you use the AIUndetectable humanizer to vary the wording, repeat the factual comparison afterwards. Natural phrasing does not verify the numbers or the recommendation. Our protected-terms guide can help you identify labels and conditions that must survive another rewrite.
Read it without the grid
Hide the source and read the paragraph aloud. Can you tell which room each price belongs to, what that price includes, and what still needs checking? Then compare every statement with the source. This second pass catches omissions that a smooth reading alone cannot reveal.
If you also publish the original HTML table, preserve its structure. The W3C WAI tables tutorial explains that header and data cells need appropriate markup so assistive technologies can convey their relationships; a caption identifies the table's topic. A prose summary is useful context, but it does not repair missing table semantics. Check the final publishing format, not only the draft in your editor.
The finished paragraph should let someone understand the same decision-relevant facts without the visual grid. It can be shorter than the table while keeping every qualification that changes what the reader can safely conclude.