All posts
Guide

Should you label examples explicitly to prevent citation errors?

By Janis Plume, Founder, Outbound Pros · 9 min read · 2026-10-27

Quick answer

Yes. If your page contains examples, hypothetical scenarios, sample numbers, or model outputs, label them explicitly and keep them visually and structurally separate from canonical facts. AI assistants often compress pages into short answers, and unlabeled examples get promoted into rules. This is most important on product, policy, documentation, and comparison pages where a mistaken citation creates support load or trust damage.

Why do unlabeled examples get cited as facts?

Because many pages are written for humans who can infer context from tone, layout, and surrounding copy. AI systems are worse at that than people assume. They are very good at extracting a crisp sentence. They are less reliable at preserving whether that sentence was a rule, a caveat, a fictional scenario, or a teaching device.

This gets worse when the example is cleaner than the real answer. Teams often write a messy qualified truth, then follow it with a neat illustrative case. The assistant lifts the neat case because it is easier to quote. You end up with a citation that is technically from your page but semantically wrong.

I see this most on pages that try to do three jobs at once: explain a concept, show an example, and push a commercial angle. That mix is not automatically bad, but it raises the burden on structure. If the page does not clearly signal what is canonical versus illustrative, the model has to guess.

A related problem is that AI crawlers do not execute JavaScript. The verified server log evidence says they fetch JS files and never run them. So if your example labels only appear after client-side rendering, or inside interactive states that depend on JavaScript, do not assume the assistant ever saw them.

If your labels depend on rendering, read our breakdown of crawler behavior.

What should you label as an example?

Label anything that is not your canonical fact but could be mistaken for one. In practice, that includes hypothetical inputs, sample outputs, illustrative pricing logic without actual pricing, fictional companies, placeholder names, made-up dates, simulated workflows, and best case snippets used only for teaching.

  • Sample scenarios, even when they look obvious to your team
  • Worked examples with fake companies or placeholder roles
  • Demonstration outputs from AI tools or calculators
  • Illustrative comparisons that simplify real edge cases
  • Template language that is not your actual policy or recommendation
  • Any sentence beginning with imagine, for example, suppose, sample, demo, or hypothetical

The rule is simple. If a sentence can stand alone and sound authoritative, label it before the model lifts it. Do not rely on a sentence after it to clean up the meaning. The cleanup often gets dropped.

How should you label examples so AI can separate them from rules?

Use blunt labels, not clever ones. Write Example, Hypothetical, Sample output, or Illustrative scenario. Put the label before the content, not after it. Keep the label in visible HTML text, close to the example, and repeated if the example spans multiple paragraphs or a table.

This is one of those cases where plain writing beats elegant writing. A human copywriter may prefer a seamless narrative. An extraction system benefits from explicit boundaries.

Weak patternStronger pattern
A paragraph starts with a confident statement, then later mentions it is just an exampleLead with Example: or Hypothetical scenario:, then state the content
Example text visually styled in a card with no textual labelVisible text label inside the card and in the first sentence
A table of sample outputs mixed with real specificationsSeparate sample table from real specs and label the table itself as illustrative
An accordion titled FAQ with a made-up use case insideA dedicated section titled Example workflow, marked as non-canonical
Client-side injected labels after page loadServer-rendered labels in the initial HTML

When examples live inside tables, headings matter. Assistants can extract facts from tables, but they may flatten rows into prose and lose nuance. If one table contains both actual facts and illustrative rows, you are creating unnecessary ambiguity. Split them.

For table extraction specifics, see this guide on comparison tables.

Where does this advice fail?

Labeling examples helps, but it is not a magic shield. If the rest of the page is messy, contradictory, or overloaded with repeated template copy, the assistant can still misread the source. Labels reduce one class of error. They do not fix weak information architecture.

It also fails when the example is too similar to the canonical fact. If your real rule and your sample rule differ by one qualifier buried halfway through a sentence, the label may not save you. In that case, rewrite the canonical rule to be shorter and more quote-ready, then place the example beneath it.

Another failure case is when teams insist on hiding the boring but important qualifier in design elements that look secondary. If the caveat is in a tab, tooltip, hover state, or collapsed panel, you are asking the model to do reconstruction work. That is a bad bet.

This advice is less important on pages that are obviously fictional or creative. It is more important on pages where readers expect source truth: docs, legal explainers, policy pages, methodology pages, and product detail pages.

Who should not follow this literally? Brands with very short pages that already contain one direct answer and no illustrations. If there is no example-like content to confuse with a fact, adding labels everywhere can make the page clunky without adding extraction clarity.

What is the cleanest page pattern for rules and examples?

Start with the canonical answer in the first paragraph. Then list qualifiers. Then place examples under a dedicated heading. That order matters because assistants often favor early, direct, self-contained statements.

  • State the rule first in one direct sentence
  • Attach the main qualifier in the same sentence or the next one
  • Use a heading that clearly separates examples from source truth
  • Repeat Example or Hypothetical inside long examples
  • Keep examples lower on the page than the canonical answer
  • Avoid mixing actual policy and illustrative language in one table or card

If your content team says this feels repetitive, good. Repetition is often what keeps the qualifier attached when the page is summarized. Elegant ambiguity is expensive once AI systems start citing your pages.

There is a second operator point here. Many teams obsess over llms.txt while leaving their visible page structure sloppy. Google states llms.txt is not used by Search, and the observed adoption rate across a large domain set is still limited, with no citation lift after controls in the cited study. So do not treat llms.txt as a substitute for making the page itself legible.

If you are deciding where to spend effort, start with the page and only then read our llms.txt analysis.

I would not overcomplicate this. Pick the pages most likely to be cited, then inspect them as if you were a lazy summarizer. Ask which sentence you would steal if you needed to answer in one line. If that sentence is an example, you have a structure problem.

  • Find pages where support, sales, or prospects often quote you back
  • Highlight every sentence that could stand alone as a definitive claim
  • Mark which of those sentences are actually examples, not source truth
  • Move or relabel any example that appears before the canonical fact
  • Server-render labels and headings so crawlers can see them without JavaScript
  • Re-test the page by asking whether the first quotable sentence is now the correct one

If you run a large site, this is less about perfection and more about risk concentration. Fix the high-stakes pages first. Product details, methods, policies, and comparisons deserve more discipline than low-consequence thought leadership.

Also, do not confuse this with outbound messaging work. If you want help with cold outreach execution, that belongs with our parent company, Outbound Pros. This site is about demand capture and AI retrieval, so the main lesson here is structural clarity on-page, not campaign execution.

The practical trade off is obvious. More explicit labels can make prose feel less slick. I still recommend them on pages where a wrong citation would cost trust. Operator rule: protect comprehension first, aesthetics second.

Common questions

Do I need to label every example on every page?

No. Prioritize pages with canonical facts, policies, product details, methods, or comparisons. Those are the pages where example confusion creates the most damage.

Is a heading alone enough to distinguish examples from facts?

Sometimes, but not always. If the example spans several paragraphs, cards, or a table, repeat the label inside the section so the context survives extraction.

Can schema solve this without changing visible copy?

Not reliably. Visible page structure still does the heavy lifting. Schema can add clarity, but it does not rescue ambiguous prose or badly mixed examples and rules.

Should examples appear above the canonical answer?

Usually no. Put the direct answer first, then qualifiers, then examples. That sequence gives assistants a better chance of lifting the right statement.

Do client-side labels count if they appear after page load?

Do not rely on them. Verified crawler evidence shows AI crawlers fetch JavaScript files and do not execute them, so server-rendered labels are the safer choice.

Last updated: 2026-10-27

Talk through your AI visibility with people who measure it

30 minutes. We will look at what assistants can actually retrieve from your site and tell you plainly what is worth fixing first.

Book a strategy call

30 minutes, no obligation. The calendar shows real availability.

Or start with the free GTM audit from Outbound Pros