Tables and comparisons
Build comparison tables with clear headers and honest gaps.
Use shared criteria, units and sources. Keep missing information explicit and preserve the comparison on narrow screens and in a linear alternative.
Free learning from Semantec SEO.
Use a table only when the comparison needs one.
A data table helps when the reader must compare the same criteria across options or inspect a structured relationship. Begin with one question, named options and shared criteria. If a short paragraph, ordered sequence or list completes the task more clearly, use that form instead. A decorative grid does not make an answer more useful.
Define the audience and common scope before writing cells. A fee for one page is not comparable with a fee for a whole site unless the difference is explicit. Keep units, time basis, inclusions and exclusions attached to the figures. Do not hide a necessary condition in prose far below the row it changes.
The table design method begins with the question, entities, criteria and approved sources, then plans the structure and context. It does not begin with a search feature wish. The visible comparison should remain useful even when a search service never extracts or displays it.
Make headers describe the relationship.
Use a meaningful caption to identify the table’s subject and scope. Name columns and rows so each cell can be interpreted with its headers. Mark header cells as th and data cells as td. For a simple grid, scope="col" and scope="row" make the intended associations explicit; more complex arrangements may need explicit header IDs and references.
The W3C table tutorial explains these relationships. Its caption guidance distinguishes identification from a longer explanation of structure. Prefer a simpler table over unnecessary merged headers that make every intersection difficult to follow.
Write a complete unit in the header or cell. “Fee” without currency, tax basis and scope is ambiguous. “Turnaround” without business or calendar days is also ambiguous. Use consistent precision; an approximate source does not justify extra decimal places. A blank cell must not silently mean zero, unavailable, included or unknown.
Compare three fictional editing proposals with explicit gaps.
This training example uses wholly invented proposals A, B and C for the same job: edit one supplied article of up to 1,500 words, with no original research. The figures and terms are stipulated teaching inputs, not live offers or measured results. Q-A, Q-B and Q-C are fictional quotation records all dated 1 October 2026. No option is a recommendation to buy a service.
| Criterion and unit | Proposal A — Q-A | Proposal B — Q-B | Proposal C — Q-C |
|---|---|---|---|
| Editing fee — EUR excluding tax | 120 EUR | 150 EUR | 130 EUR |
| Setup fee — EUR excluding tax | Not provided | 0 EUR, explicitly waived | Not applicable: no setup service offered |
| Included revision rounds — count | 1 | 2 | 1 |
| Stated turnaround — business days after accepted brief | 5 | 4 | Not provided |
| Delivery format — declared format | Editable document | Editable document and PDF | Editable document |
| Live collaboration access — quoted condition | Not provided | Named client account | Not applicable: file delivery only |
| Measured error reduction — study outcome | Not measured | Not measured | Not measured |
Read Proposal B’s editing fee as 150 EUR excluding tax for the stipulated article scope. Its zero setup fee is supported by an explicit fictional waiver. Proposal A’s missing setup fee is unknown, not zero. Proposal C’s “not applicable” describes a service boundary, not a hidden free setup offer. None of the proposals supplies an error-reduction study.
The source record for every A cell is Q-A, for every B cell Q-B, and for every C cell Q-C. These records are deliberately finite: the comparison must not infer a faster process, better quality, broader rights or lower final bill from facts they do not provide. In a real comparison, link or identify the permitted source and date beside the relevant assertions.
Explain missing states before choosing an option.
Use distinct words for distinct gaps. “Not provided” means the reviewed source omits the fact. “Not applicable” means the criterion does not belong to that option under the stated scope. “Not measured” means no relevant measurement supports the outcome. “Unavailable” means a source affirmatively says the feature is absent. Do not replace all these states with a dash.
In the fictional scenario, a reader needing a four-business-day turnaround has support only from B’s stated term; C’s missing term cannot be assumed faster or slower. A reader with a strict total-cost ceiling cannot finish the A comparison until its setup fee is clarified. A declared delivery format does not establish accessibility, and more revision rounds do not by themselves establish better editing.
Do not calculate a winner score that treats unknowns as zero or rewards an option for missing data. Record the unresolved condition and next source question. A conditional recommendation is more honest than a precise-looking total built from incompatible or absent inputs.
Provide a readable alternative to the wide grid.
A narrow screen should preserve the relation between criterion and option without making the whole page overflow. A genuinely wide table may use controlled local scrolling with a visible affordance, keyboard access and clear labeling. Inspect it on the delivered page. A swipe-only region without a keyboard path is insufficient.
An option-by-option text alternative can help readers who prefer a linear view. Preserve every criterion and gap, not just the favorable fields. The following uses the same fictional source records and facts as the table.
Proposal A — Q-A
Editing fee: 120 EUR excluding tax. Setup fee: not provided. Included revisions: 1 round. Stated turnaround: 5 business days after accepted brief. Delivery: editable document. Live collaboration access: not provided. Measured error reduction: not measured.
Proposal B — Q-B
Editing fee: 150 EUR excluding tax. Setup fee: 0 EUR, explicitly waived. Included revisions: 2 rounds. Stated turnaround: 4 business days after accepted brief. Delivery: editable document and PDF. Live collaboration access: named client account. Measured error reduction: not measured.
Proposal C — Q-C
Editing fee: 130 EUR excluding tax. Setup fee: not applicable because no setup service is offered. Included revisions: 1 round. Stated turnaround: not provided. Delivery: editable document. Live collaboration access: not applicable because delivery is by file only. Measured error reduction: not measured.
The alternative changes presentation, not evidence. It should come from the same maintained record so a later update cannot leave the grid and linear account disagreeing. Do not visually hide material differences merely to achieve a compact mobile layout.
Review the table’s sources and its surrounding explanation.
Check each intersection against its source, date, market, version and scope. Verify unit conversions and retain the original basis when a conversion is supplied. Record uncertainty and remove stale figures when no current source supports them. A table is not exempt from ordinary factual review because its cells are short.
Keep setup, takeaway and longer qualifications outside dense cells while leaving the relevant limitation easy to find. A reader should be able to explain what is compared, under which conditions, what is missing and what would change the decision. Avoid a large winner badge that contradicts the missing-evidence record.
Check headings and table semantics, zoom, reading order, local scrolling, keyboard focus and the linear alternative. The W3C relationship criterion supports making conveyed structure available beyond appearance. A caption and th tags are useful ingredients, not a complete accessibility certification.
Maintain the comparison without promising a search display.
Keep a record of the question, compared entities, criteria, units, sources, missing states, accepted recommendation, owner and update trigger. Reopen it when prices, scope, availability, source documents or reader conditions change. If options stop being comparable, revise the question or split the table instead of silently changing the criteria for one column.
Google’s content usefulness questions provide a general editorial check. Neither this table structure nor a chosen row count guarantees extraction, rankings or another search appearance. The useful outcome under the publisher’s control is a clear, sourced comparison with honest gaps.
Return to the visual meaning and usable pages hub for related hierarchy and explanation work. The learning overview connects tables with source ownership and content organized around the reader’s task.