Title, description, canonical: what a citation is attached to

A citation is three fields

Look at how ChatGPT search, Perplexity or Google's AI answers show a source: a short label, sometimes a line of context, and a link. The label comes from your <title> (or the Open Graph title). The context often comes from your meta description or the first paragraph. The link is your canonical URL. These three fields are the entire interface between your page and the reader who might click through.

Pages with a missing title get labelled by their URL slug or by whatever the engine invents. Pages with several URLs and no canonical get their credit split across duplicates, or attributed to the tracking-parameter version. None of this is visible in a content audit that looks at the body.

What Qomvia checks

Title, description and canonical URL in the methodology is worth 6 points, two per field. A title needs at least 10 characters, a meta description at least 50, and the canonical link just has to exist. The result lists what is missing, so the fix is a single line per field.

The check runs on the most readable page the crawl fetched, a content page where one is reachable, otherwise the homepage. The typical failure differs by page type: homepages usually have a title and no description, content pages usually have both and no canonical, so check both kinds.

Writing the three fields

• Title: the subject, then the brand. 'Refund policy for Swiss orders | Brand', under about 60 characters so it is not cut. Do not put volatile values in it: a score, a date or a price in the title churns the indexed label every time it changes and gets cached in the old state.
• Description: the answer in one sentence. 50 to 160 characters. Say what the page states, not what it is about: 'Unopened items can be returned within 30 days; shipping is refunded for faulty goods.' That sentence is what an engine can show as context without reading further.
• Canonical: absolute, self-referencing, one per page. https:// scheme, the host you want credited, no tracking parameters. Every variant (?utm=, trailing slash, m. subdomain) points at the same canonical.

<title>Refund policy for Swiss orders | Brand</title>
<meta name="description" content="Unopened items can be returned within 30 days. Shipping is refunded for faulty goods. Returns are free within Switzerland.">
<link rel="canonical" href="https://www.brand.ch/help/refunds">
<meta property="og:title" content="Refund policy for Swiss orders">
<meta property="og:description" content="Unopened items can be returned within 30 days. Shipping is refunded for faulty goods.">

Titles and the temptation to be clever

A title is not a headline. Headlines are for people who are already on the page; titles are for readers deciding whether to arrive, and for machines deciding what to call you. 'The truth about returns' tells neither what the page contains. Qomvia's own site pages moved the score out of the title for exactly this reason: a stable label in the index, the changing number in the description, the Open Graph image and the structured data.

Dates belong in structured data too, not in the title. The freshness article covers where they go.

Questions

Do AI engines use the meta description?

They use whatever short text is available to describe a source, and the description is the field designed for that. Where it is missing they take the first paragraph, which is often a greeting or a cookie notice.

Is a canonical needed if there is only one URL for the page?

Yes. Tracking parameters, campaign links and shared URLs with fragments all create variants you did not publish. A self-referencing canonical collapses them.

Should the title match the H1?

They should say the same thing; they do not have to be identical. The H1 can be longer and read as a headline, the title should be shorter and read as a label.