The 2026 Q4 Stupidity Index
Ten SEO mistakes we keep running into in late 2026, each with its evidence and its fix. They're in no scientific order. Number one is just the one that annoyed me most.
1. Fixing SEO with a JavaScript pixel
The mistake: letting a tool rewrite your titles, descriptions and structured data in the browser instead of in your HTML.
The evidence: we tested five sites running an SEO platform's pixel in October 2026. On all three with live changes, none of the fixes were in the HTML that AI crawlers receive. The full test.
The fix: put the changes in your templates and turn the pixel off.
2. Blocking AI search to stop AI training
The mistake: disallowing every crawler a company runs when you only objected to training.
The evidence: OpenAI, Anthropic and Perplexity each publish separate crawler names for training and for search. OpenAI says blocking GPTBot doesn't affect ChatGPT search; blocking OAI-SearchBot does. The crawler map.
The fix: block the training names, allow the search names, if that's what you want.
3. Selling FAQ schema as a search win
The mistake: adding FAQPage markup to get the expandable questions under your listing.
The evidence: since August 2023, Google shows FAQ rich results only for well-known, authoritative government and health sites. What changed.
The fix: write Q&A for readers and spend markup effort on types that still produce a result for you.
4. Meta keywords, still
The mistake: maintaining a meta keywords tag.
The evidence: Google said in 2009 that it doesn't use the keywords meta tag in web ranking. The pixel in item 1 was still adding them in 2026.
The fix: delete it. All it does now is tell competitors which keywords you're targeting.
5. Sitemaps where every page changed today
The mistake: stamping every URL's lastmod with the time the sitemap was generated, and tuning priority and changefreq.
The evidence: Google ignores priority and changefreq and uses lastmod only when it's consistently accurate. The details.
The fix: real lastmod dates, canonical URLs only, declared in robots.txt.
6. Homepages with no links in the HTML
The mistake: building the whole page in the browser.
The evidence: in the same test as item 1, a car broker's raw homepage HTML had no H1 and zero links. Crawlers that don't run JavaScript saw an empty site. How to check yours.
The fix: render on the server or at build time, and use real links.
7. Four homepages, chained together
The mistake: redirect rules stacked over the years, so http://www takes two or three hops to reach the real homepage.
The evidence: Google follows up to 10 redirect hops, so it gets there, but every hop is wasted time for every visitor and crawler. The fix in four steps.
The fix: one 301 from each version to the final address.
8. Structured data that makes claims you can't back
The mistake: letting a plugin or tool fill every schema field it can.
The evidence: the pixel in item 1 injected favicon-service logos and a Creative Commons license for the sites' photos, including a stock photo. Google's guidelines say markup must match visible content, and misleading markup can earn a manual action. Examples.
The fix: read your structured data like a contract and delete what isn't true.
9. Link previews nobody checked
The mistake: never looking at what your page looks like when someone shares it.
The evidence: preview bots generally read Open Graph tags from the raw HTML, so missing, relative or JavaScript-added tags produce a blank card. What to check.
The fix: page-specific title, description and image tags, with absolute URLs, in the server's HTML.
10. Publishing AI pages by the thousand
The mistake: generating pages at scale to catch every keyword variation.
The evidence: Google's spam policies name this scaled content abuse: many pages made mainly to manipulate rankings, however they're produced.
The fix: fewer pages, each worth reading. See GEO without the snake oil for what earns citations instead.
How we built this list
This first edition draws on one test we ran ourselves (five sites, home pages only, October 29, 2026; details and limits in the pixel article) and on the search engines' and AI companies' own documentation, linked in each item. It isn't a large-scale scan study. Later editions will add scan data, with the sample and dates published.