How to revise a scientific figure legend without changing the figure facts
By GPTHumanizer’s AI operations assistant · Writing guide
Published · Updated
To revise a scientific figure legend, keep the figure and its underlying record beside the draft. First correct what the caption says about each panel, plotted point, unit and error bar. Then improve the wording. A fluent sentence can still describe the wrong panel or turn repeated measurements into independent samples. This guide uses an invented figure record to show the difference between correcting those facts and making a checked passage easier to read.
1. Check the caption requirements
Use the current instructions for the journal you are submitting to. A figure caption may include a label, title and explanatory legend, but journals differ in terminology and required formatting. Keep that formatting check separate from the question of whether your caption describes the correct figure.
PLOS ONE’s figure instructions require a figure label and title; an explanatory legend is optional. Its guidance calls for concise panel explanations and definitions of nonstandard symbols and abbreviations. PLOS specifies a title of no more than 15 words and says formatting requirements are waived until provisional acceptance. Those are PLOS instructions, not a universal caption template.
Psychological Science’s manuscript guidelines say captions should describe the figure rather than summarize results, identify panels, explain error bars and expand abbreviations. Its lowercase panel-letter convention belongs to that journal. Follow the actual destination’s instructions instead of transferring that convention to every figure.
The worked example below uses uppercase A and B because those are the labels in its fictional figure record. It is an editing exercise, not a submission-ready format for either journal.
2. Build a figure record
All experimental details below are invented for this exercise. This fictional example was created for this guide; it is not customer text, a real experiment or recorded Humanizer output. The table describes an imagined figure; no actual graph or measured dataset is supplied.
Swipe across to read all columns.
| Item | Record for this exercise |
|---|---|
| Figure and timing | Figure 1 contains panels A and B. Both report measurements taken 14 days after the start of the two light conditions. |
| Panel and unit mapping | A: shoot height in centimetres (cm). B: leaf count per seedling. Neither panel shows a change from baseline. |
| Experimental units | 24 pots in total, assigned at the pot level: 12 to low light and 12 to moderate light. The pot is the experimental unit; n is 12 pots per condition. |
| Seedlings and plotted points | Each pot contains three measured seedlings. A point in either panel is the mean of that panel’s measurements from those three seedlings within one pot. There are 12 points per condition in each panel. |
| Relationship between panels | The same 24 pots and their seedlings supply both panels. The panels are two measurements of the same experimental units, not additional independent samples. |
| Symbols | Filled circles identify low light; open circles identify moderate light in both panels. |
| Summary marks and error bars | Horizontal marks show the mean of the 12 pot means within each condition. Bars extend one standard deviation (SD) above and below that mean, calculated across those 12 pot means. |
| Statistical comparisons | No statistical comparison between conditions was performed. The record supplies no p-value, confidence interval or test result. |
For a real caption, establish these entries from the figure, analysis record and responsible author. A draft caption is not its own evidence. If you cannot tell what one point or error bar represents, ask for that information before editing its description.
3. Compare the complete legend drafts
The first draft below deliberately disagrees with the record. The revision corrects those discrepancies before changing the prose. Its extra words explain the unit of analysis and the bars; a shorter caption would not be better if it kept the wrong facts.
Before and after: an editing example
Fictional example created for this guide; not customer text or recorded Humanizer output. Complete legend: 116 words before; 130 words after.
The before-text intentionally conflicts with the fictional record. The after-text is an editorial correction using that record, not a demonstration that Humanizer identified or repaired the errors.
Original draft
Figure 1. The higher-light condition significantly improves seedling growth. Panel A shows leaf count and panel B shows shoot height after 14 days in the two conditions. The figure contains 36 independent observations per condition, because three seedlings were measured in each of 12 pots. Every dot is an individual seedling, and the same seedlings appear in both panels. Filled circles identify moderate light; open circles identify low light. The horizontal marks show the average in each condition. Error bars are 95% confidence intervals, demonstrating that the difference between conditions is statistically significant. Shoot height is expressed in millimetres. The figure therefore shows that moderate light produces better seedlings, with the effect confirmed in both panels.
Edited example
Figure 1. Seedling measurements after 14 days under two light conditions. Panel A shows shoot height in centimetres (cm); panel B shows leaf count per seedling. The same 24 pots are represented in both panels, with 12 pots assigned to low light and 12 to moderate light. Each point is the mean of measurements from three seedlings in one pot, not an individual seedling measurement. The pot is the experimental unit, so n = 12 pots per condition in each panel. Filled circles indicate low light; open circles indicate moderate light. Horizontal marks show the mean of the 12 pot means for each condition. Error bars extend one standard deviation (SD) above and below that group mean, calculated across the 12 pot means. No statistical comparison between conditions was performed.
The revision restores the A/B and symbol mappings, centimetre unit, pot-level n and SD definition. It removes unsupported significance and benefit claims. These corrections depend on the supplied record; an optional language rewrite comes only after this factual check.
The edited title identifies what is displayed without making a significance or benefit claim. The body names the panels in the order shown and explains the point, sample count, symbols and bars.
4. Audit the changes against the record
Swipe across to read all columns.
| Draft problem | Correction from the record | Why it matters |
|---|---|---|
| Panels and height unit are reversed or wrong | A is height in cm; B is leaf count. | Preserving A, B and a number somewhere in the text is insufficient: each must remain attached to the right measurement. |
| 36 independent observations and individual-seedling dots | Each point is a three-seedling pot mean; n = 12 pots per condition. | There are 36 measured seedlings per condition, but three share one experimental unit. Rewording must not promote those seedlings to 36 independent units. |
| Filled and open circles identify the wrong groups | Filled is low light; open is moderate light. | A fluent description with swapped symbols directs the reader to the wrong plotted group. |
| 95% confidence intervals and claimed significance | Bars are ±1 SD across 12 pot means; no comparison test was performed. | A standard deviation, a standard error and a confidence interval are not interchangeable labels. The caption cannot supply an analysis that was not performed. |
| Two panels treated as proof of improvement | Describe height and leaf count at day 14 without a benefit claim. | These panels use the same pots. Two measurements do not create another independent sample, and the record provides no baseline change or tested effect. |
The edit preserves the 14-day timing and both light-condition names. It also keeps the same pot count in each panel. Check relationships of this kind, not just whether the original digits survived.
5. Resolve missing facts before polishing
Make a separate author-check note for anything the record does not settle: what n counts, whether the same units appear in different panels, what the bars represent, or which test actually produced a displayed annotation. Keep the note outside the proposed caption until it is resolved. Do not use a likely convention to fill the gap.
For example, knowing that 36 seedlings were measured does not tell you the experimental unit without the design record. Likewise, the appearance of a bar does not establish whether it is SD, standard error or a confidence interval. Ask for the recorded definition; do not infer it from the picture.
Once those facts are confirmed, read the caption while pointing to each corresponding part of the figure. Check the figure number against its in-text citation too. If you discover a mismatch, correct the figure or caption from the authoritative record before calling the prose final.
6. Try a checked passage in Academic mode
For a wording pass, first prepare a factually checked legend such as the edited 130-word example. Paste your passage into the English Humanizer editor and choose Academic. The tool receives text: it does not inspect your figure, determine the experimental unit, correct a panel mapping or verify a statistical calculation.
Rewriting requires sign-in and sufficient available input words. New accounts have a one-time 300-word allowance shared with Essay Writer, subject to eligibility and regional availability. Opening this link only opens the editor; it does not load the example or submit a rewrite.
Open the English Humanizer editor
After a suggestion returns, compare every panel label, symbol, unit, count, abbreviation and bar definition with the checked version. Keep the original if a smoother sentence changes one of those relationships. For real research material, follow the applicable AI-assistance and sharing rules before uploading it.
For the final wording check, use the academic claim-by-claim review checklist.
Your final review
- The caption refers to the correct figure number and every panel label.
- Each measurement retains its unit, timing and displayed group.
- The meaning of one point and the unit counted by n match the study record.
- Repeated measurements or panels have not become extra independent samples.
- Symbols and abbreviations are defined consistently with the figure.
- The error-bar definition and any statistical statement match the actual analysis record.
- Missing information has been resolved before the legend is treated as final.
- Any optional wording suggestion has been checked against the verified caption.