Guide · Practitioner
Data Visualization and Charts for Citations
Charts get cited when their key numbers also exist as text: a one-line takeaway, a caption, an HTML data table, alt text, and a stated method and source. Google says it uses alt text, captions and surrounding page content to understand images (Google). W3C classes charts as complex images that need a long description in addition to short alt text (W3C).
Facts are from Google Search Central and W3C WAI, accessed 30 Oct 2026. Recommendations marked "our view" are interpretation.
On this page
- Why are charts hard to cite?
- What should accompany every chart?
- How do you write alt text for charts?
- Why use HTML data tables?
- How do you state methods and sources?
- Which chart types work best?
- Should charts be images or code?
- What about filenames and image SEO?
- What's a worked example?
- How do you measure whether charts get cited?
- Which pitfalls should you watch for?
- How should you handle uncertainty?
- How do you design charts people can describe?
- Should you publish the raw data?
- How do you handle updated data?
- Checklist
- From MentionX data: cited page types and engine behaviour
- Worked example: Ledgerline makes its research chart citable
- How do you write a chart caption?
- How do you title a chart?
- How do you design charts that are easy to describe?
- How do you make charts accessible?
- How do you cite your own data correctly?
- How should you handle chart updates?
- How do you promote chart-based research?
- What file names and metadata help?
- Who should check charts before publishing?
- Common questions
- Related
- Sources
Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.
Why are charts hard to cite?
Our view: a chart is pixels. Some systems can analyse images, but reading exact values from a chart is error-prone, and many retrieval pipelines rely mainly on text. If your key finding lives only inside an image, an engine may miss it or misread it, and a human journalist can't copy the number accurately. The fix is simple: say it in words and tables as well.
What should accompany every chart?
| Element | Example (fictional) |
|---|---|
| Takeaway sentence | "Invoice approval time fell from 6.1 to 2.3 days after automation." |
| Title | "Median approval time, 2025–2026" |
| Caption | Source, sample, date |
| Alt text | "Line chart: median approval time fell from 6.1 to 2.3 days" |
| Data table | HTML table with the values |
| Method note | How data was collected |
How do you write alt text for charts?
Keep alt text short: chart type and main finding. Put the full data in a nearby table or description, as W3C recommends for complex images (W3C). Google advises descriptive, non-stuffed alt text (Google). See Image alt text for multimodal AI search.
Why use HTML data tables?
Our view: a table is the most quotable form of numbers. Use real HTML tables with header cells, as W3C's tables tutorial describes (W3C), not images of tables. Readers can copy values, screen readers can navigate them, and engines can read them as text.
How do you state methods and sources?
- Who collected the data and when
- Sample size and how it was chosen
- Definitions (what "approval time" means)
- Any exclusions or caveats
- Link to the primary source if third-party
Google's helpful content guidance values original research with clear sourcing (Google). Our view: a chart without a method is an assertion, not evidence.
Which chart types work best?
| Data | Chart |
|---|---|
| Change over time | Line |
| Comparing categories | Bar |
| Parts of a whole (few parts) | Stacked bar |
| Distribution | Histogram |
| Relationship | Scatter |
Our view: simpler charts are easier to describe in one sentence. If you can't summarise a chart in one sentence, consider splitting it.
Should charts be images or code?
Either can work. SVG or HTML/JS charts can include text labels in the page; PNG/JPG need alt text and tables. Our view: whatever the format, keep the data table in HTML. Check that script-rendered charts don't hide the table from crawlers.
What about filenames and image SEO?
Use descriptive filenames, for example approval-time-2025-2026.png, and place charts near related text, as Google recommends (Google). Google also lists image sitemaps and structured data among ways to help it find images.
What's a worked example?
Fictional example: a logistics company publishes an annual delivery-delay report. Previously the key chart was a PNG in a PDF. Now each chart sits on an HTML page with a takeaway sentence above it, a caption with sample size and date, short alt text, a data table beneath and a downloadable CSV. Journalists and engines quote the takeaway numbers directly.
How do you measure whether charts get cited?
- Search for your key figures in quotes to find mentions
- Prompt tests asking about the statistic
- Links to the data page
Which pitfalls should you watch for?
- Key numbers only in the image
- Charts in PDFs only
- No date or sample size
- Truncated axes that exaggerate change
- Decorative 3D charts that distort values
How should you handle uncertainty?
State sample sizes and, where relevant, ranges or margins of error. Our view: note when differences are small or the sample is limited. Overstated precision undermines trust when others check your numbers, and honest caveats make citations more likely to repeat your finding accurately.
How do you design charts people can describe?
Our view: one message per chart, a title that states the finding rather than just the metric, labelled axes starting at zero for bar charts, and direct labels instead of a legend where possible. Use colour sparingly and don't rely on colour alone to distinguish series, which also helps colour-blind readers. A chart designed so a reader can say its point in one sentence is a chart that others can quote accurately.
Should you publish the raw data?
Where you can, yes. A downloadable CSV alongside the HTML table lets journalists, researchers and analysts check and reuse your numbers, and each reuse is a chance for a citation back to you. Our view: state the licence, for example whether attribution is required, and include a "how to cite" line with the page title, organisation and date.
How do you handle updated data?
When you update a chart with new figures, keep earlier versions available or note what changed and when. Our view: people may already have cited the old numbers; an archive or change note lets them see the difference rather than assuming an error.
Checklist
- Takeaway sentence per chart
- Caption with source, sample, date
- Short alt text
- HTML data table
- Method note
- Descriptive filename
- Not only in PDF
From MentionX data: cited page types and engine behaviour
Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.
This section itself shows the approach: the data is in text tables rather than images. In the MentionX sample, articles are cited most often:
| Cited URL type | Share of 9,750 citations |
|---|---|
| Blog or article | 19.6% |
| Homepage | 10.0% |
| Review page | 2.7% |
| Docs | 2.6% |
| Pricing | 1.9% |
| Wikipedia | 1.6% |
Engines differ in how often they cite anything:
| Engine | Brand mentioned | Answers with no citations |
|---|---|---|
| Gemini | 47.3% | 50.7% |
| OpenAI | 42.9% | 30.2% |
| Perplexity | 31.5% | 4.1% |
| AI Mode | 31.3% | 12.6% |
| Copilot | 26.0% | 47.0% |
| DeepSeek | 24.0% | 99.0% |
Our view: a chart inside an article is only as citable as the text around it. Engines that cite often are citing text, so the finding needs to be in words and a table, not only in pixels.
Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.
Worked example: Ledgerline makes its research chart citable
Ledgerline is a fictional accounts-payable tool. Figures are illustrative.
Ledgerline publishes a report on invoice processing times, with a bar chart comparing teams that automate approvals and teams that don't. The chart is a PNG with no alt text, and the article only says "see the chart below".
The research lead adds a caption stating the main finding in one sentence, with the sample size and date. The alt text summarises the chart. Below it, an HTML table lists the values for each bar. A short methods note explains how the data was collected. The raw data is published as a CSV. All figures in the report are hypothetical for this example.
How do you write a chart caption?
State the main finding in one sentence, including the key number, the group or time period and the source. Add the sample size and date if relevant. Keep it plain: "Teams using automated approvals processed invoices faster than teams using email approvals, in a survey of finance teams in 2026" is more useful than "Approval speed by method".
How do you title a chart?
Use a title that describes what the chart shows, with units and time period. A title can state the finding, but the caption should still give the numbers. Avoid titles that are only labels, such as "Figure 3".
How do you design charts that are easy to describe?
Keep charts simple: one message per chart, clear labels, readable fonts and a limited number of colours. Label data directly rather than relying on a legend where possible. Avoid 3D effects and decorative elements. If a chart is hard to describe in a sentence, it may be trying to show too much.
How do you make charts accessible?
Use sufficient colour contrast and don't rely on colour alone to show differences. Add patterns or labels for people with colour blindness. Provide the data in a table for screen reader users. These steps help everyone, including engines.
How do you cite your own data correctly?
State where the data comes from, when it was collected and how. If you use third-party data, link to the original source and follow its licence. If figures are estimates or modelled, say so. Readers and engines trust charts more when the method is visible.
How should you handle chart updates?
When data changes, update the chart, caption, table and text together. Add a note with the update date and what changed. Keep old versions available if people have cited them, or explain where the new figures differ.
How do you promote chart-based research?
Share the key finding and a link to the page, not only the image. Offer journalists the data table and methods note. Make it easy to embed the chart with a credit link. Our view: the page with the text, table and method is the asset worth promoting; the chart is how people notice it.
What file names and metadata help?
Name chart files descriptively, such as the topic and year, rather than generic names. Add a title and description in the image metadata if your tools support it. These help image search and keep your files organised.
Who should check charts before publishing?
Ask someone who didn't make the chart to read the caption and table and say what the chart shows. If their answer matches your finding, the chart is clear. Ask an analyst to check the numbers against the source data.
Common questions
Can engines read text inside chart images? Some can, but it's unreliable; put the key text in HTML.
Should I use SVG or PNG for charts? Either can work; what matters most is the text and table around the chart.
Do I need a table for every chart? For charts with important data, yes; for simple illustrations, a clear caption may be enough.
Related
- Image alt text for multimodal AI search
- Original research as a GEO moat
Sources
- Google, Image SEO best practices; Creating helpful content
- W3C WAI, Images tutorial; Tables tutorial
All accessed 30 Oct 2026.
Questions
Can AI engines read charts?
Some systems analyse images, but text is the dependable format. Google says it uses alt text, captions and page content to understand images. Put key numbers in text and tables too.
What's the minimum for an accessible chart?
Short alt text naming the chart and its main point, plus a longer description or data table nearby. W3C classes charts as complex images.
Should I publish the underlying data?
Where possible, yes, as an HTML table or downloadable file with method and date. It lets others check and cite your numbers.