SEO meta tags that actually matter in 2026 (and the ones you can skip)

The four tags that carry real weight

Everything else is either automatic or noise. These four earn their place on every page:

1. <title>. Still the strongest on-page ranking signal. 50-60 characters, front-load the primary keyword, make it click-worthy (Google now rewrites titles it thinks are misleading, so don't stuff).

2. <meta name="description">. Doesn't directly affect ranking, but it's the snippet under your result. A good description raises click-through rate, which affects ranking indirectly. 140-160 characters.

3. <link rel="canonical">. Tells Google which version of duplicate URLs to index. Skip this and the same page indexed at /foo, /foo/, /foo?ref=abc splits your ranking signals three ways.

4. Open Graph tags (og:title, og:description, og:image, og:url). Not for Google — for every share on Slack, Discord, LinkedIn, iMessage. A missing OG image is a link with no thumbnail, and it visibly reduces click-through.

Tags you can skip in 2026

• <meta name="keywords"> — Google publicly dropped support in 2009. Every "SEO checklist" that still includes it is copy-pasted from 2005.
• <meta name="author"> — no ranking impact. Use <link rel="author"> or JSON-LD Person schema if you care.
• <meta name="robots" content="index,follow"> — this is the default. Only set it if you're changing the default (noindex, nofollow).
• <meta http-equiv="X-UA-Compatible"> — targeted IE 8-10. IE is dead.
• Twitter card tags if you also have OG tags. Twitter (X) falls back to OG. twitter:card is the only one worth adding, and only to force summary_large_image.

The tags AI crawlers actually check

New in 2024-2026, largely undocumented:

robots.txt per-bot allowlist. GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot — each has its own user-agent string, and each honors an explicit Allow or Disallow rule for its own name. If you want to appear in AI answers, allow them. If you want to block training, disallow.

Meaningful structure. AI crawlers extract facts from H2/H3 hierarchy and bullet lists more than paragraph prose. The same structure that helps human skimmers helps AI extraction. See the robots.txt on this site as an example — GPTBot, ClaudeBot, and PerplexityBot are explicitly allowed.

JSON-LD structured data. The strongest hint about what a page is. SoftwareApplication for a tool page, FAQPage for FAQs, HowTo for tutorials, Article for blog posts. Google's rich results build on it, and every AI crawler that hits your page reads it.

Open Graph: the details that trip people up

Image size. 1200×630 is the safe standard. Slack renders any size fine; LinkedIn crops aggressively below 1200 width; Discord uses whatever you send.

Image format. PNG or JPG. Not SVG — many social platforms reject it silently. Ship a PNG even if you have an SVG source; the PNG unfurl works everywhere.

Absolute URLs. og:image must be an absolute URL (https://example.com/og.png), not a relative path. Cache-buster query strings often break unfurlers; use immutable filenames instead.

og:url matches the canonical. If they diverge, some scrapers pick one and others pick the other, splitting share metrics.

The Google-preview double-check

Every meta-tag guide tells you to preview in Google. Nobody says how to actually do it — Google's search-appearance tool needs verified site ownership. The practical shortcut:

• Title should be under 580 pixels wide (roughly 60 characters, more if narrow letters).
• Description should be under about 160 characters for desktop, 120 for mobile.
• URL slug should be short, lowercase, hyphen-separated. Long slugs get truncated in the display.

Meta Tag Generator shows a Google-style preview with character-count warnings — the equivalent of eyeballing the SERP without waiting for Google to re-index.

Structured data is where the leverage is

The single highest-impact addition most sites are missing: a FAQPage JSON-LD block on any page that answers questions. Google renders these as expandable FAQ blocks in results, taking multiple SERP slots for one page.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    { "@type": "Question", "name": "What is X?",
      "acceptedAnswer": { "@type": "Answer", "text": "X is …" } }
  ]
}
</script>

Every tool page on this site ships FAQPage schema for that reason. Add it to yours.

The five-minute audit

For any existing page:

1. Title 50-60 chars, front-loaded keyword?
2. Description 140-160 chars, action-oriented?
3. Canonical URL set, absolute?
4. OG image is a real PNG 1200×630 at an absolute URL?
5. Structured data present — at minimum WebSite + Organization on the homepage, Article on blog posts, FAQPage on anything Q&A shaped?

If yes to all five, you're doing better than most of the web.

Related tools:

• Slugify — for clean URL slugs.
• Markdown to HTML — for the content itself.