7 GEO Mistakes That Kill AI Visibility (and How to Fix Them)

Short answer: on the seven sites we measured, lost visibility was not a writing problem. It came from pages Google has decided not to index, sitemaps it stopped reading, duplicate URLs, and content that repeated what every other page already says. Fix those in order: get the page indexed, give it one canonical URL, then add something no competing page has. Rewording the text does not help.

Everything below was measured, not estimated. On September 14, 2026 we ran Google Search Console’s URL Inspection on every sitemap URL across seven websites we operate (one large directory was sampled at 400 of 2,291 URLs), and pulled 16 months of Search Console and 90 days of GA4 data for each. Where we cite outside research, it is linked.

Editor’s note: an earlier version of this article cited statistics we could not source. We removed all of them and rebuilt the page from the measurements below.

What we found across seven sites

  • 419 of 1,482 inspected URLs (28%) were not in Google’s index. 185 were “Discovered – currently not indexed” (Google knows the URL and chose not to fetch it), 57 were “Crawled – currently not indexed” (fetched, then rejected), and 120 were unknown to Google entirely.
  • More posts earned less. On one site, 12 posts published in May earned 210 clicks. The 62 posts published the following month earned 14 clicks combined.
  • AI assistants sent almost nothing. Across all seven sites, ChatGPT, Claude, Gemini and Perplexity referred 26 of 3,842 sessions in 90 days, under 1%.

Mistake 1: Publishing more pages than the site can make useful

Google’s spam policy defines scaled content abuse as “many pages generated for the primary purpose of manipulating search rankings and not helping users,” and lists “using generative AI tools or other similar tools to generate many pages without adding value for users” as an example (Google Search spam policies). The policy judges value, not authorship.

The pattern shows up in outside data too. Lily Ray tracked 220+ sites named as customers of AI content platforms: 54% lost at least 30% of their peak organic traffic, and 22% lost at least 75%, usually within a year of the peak (Lily Ray, May 2026). Her figures are third-party estimates, so treat them as a pattern, not a precise rate.

Fix: before publishing, check whether the page contains anything the top five results for its query do not. If it doesn’t, don’t publish it. For existing pages, merge near-duplicates into the strongest one with a 301 redirect.

Mistake 2: Claiming research you did not do

“We analyzed 200 implementations” and “85% of failures” read as authority, but a model citing your page, or a buyer checking it, cannot verify a number with no source. Invented experience is the opposite of the experience signal it imitates. This page made that mistake.

Fix: every statistic gets a link or a stated method. First-party data, even small, is the one thing competitors cannot copy: your own Search Console exports, support tickets, sales-call questions, before-and-after screenshots.

Mistake 3: Treating “Discovered – currently not indexed” as a waiting period

A page can sit in that state for months. This article is an example: it had earned 2,063 Search Console impressions, then dropped to “Discovered – currently not indexed” and stopped appearing entirely.

Automated rewording does not change Google’s assessment. The same spam policy names “automated transformations like synonymizing, translating, or other obfuscation techniques” as part of the abuse pattern.

Fix: for each page stuck in that state, choose one: add information the page lacks and link to it from pages Google already indexes, merge it into a stronger page, or noindex it. Don’t leave hundreds of them in the sitemap.

Mistake 4: A sitemap Google stopped reading

One of our sites had not had its sitemap downloaded by Google in eight weeks. Since Google last processed it, the sitemap had grown from 209 to 279 URLs, and 84 URLs on the site were unknown to Google.

Fix: Search Console → Sitemaps shows “Last read” for each sitemap. If it is more than a week old on a site that publishes regularly, resubmit it. The Search Console API can check and resubmit on a schedule. Note that Google’s Indexing API is not a shortcut: it “can only be used to crawl pages with either JobPosting or BroadcastEvent embedded in a VideoObject” (Google Indexing API docs).

Mistake 5: A lastmod date that changes on every request

Many frameworks generate sitemaps with the current time as every page’s last-modified date. On one of our sites, 79 URLs claimed to have changed at the moment each request was made. Google says it uses lastmod only “if it’s consistently and verifiably accurate” (Google sitemap documentation). A date that is always “now” teaches it to ignore yours.

Fix: set lastmod from the content’s real update time, or leave it out. Google also ignores priority and changefreq.

Mistake 6: Two URLs for the same thing

An ecommerce site we run serves each product at two addresses: a detailed product page and a store stub (for one product we measured, 492 words versus 81). For 11 products, Google picked the stub as the canonical version, so the thin page is the one eligible to appear.

Fix: in URL Inspection, compare “User-declared canonical” with “Google-selected canonical.” Where they differ, make one URL redirect or declare the other as canonical, and keep only that URL in the sitemap.

Mistake 7: Measuring rankings but not the answer engines

Search Console does not report whether ChatGPT or Perplexity cited you. GA4 does record their visits: filter Traffic acquisition by session source for chatgpt.com, perplexity.ai, gemini.google.com and claude.ai. That is how we found 26 referrals out of 3,842 sessions.

Fix: track AI referral sessions monthly alongside Search Console clicks, and run your target questions in each assistant to see which sources it cites instead of you.

Once a tracker flags a visibility problem, fix it in this order

  1. Indexing. Inspect the URL. If it isn’t indexed, nothing else matters yet.
  2. Canonical. Confirm Google selected the URL you intended.
  3. Discovery. Confirm the sitemap is being read and the page is linked from indexed pages.
  4. Accuracy. Correct anything the assistant states wrongly about you on your own pages first: prices, services, locations, policies.
  5. Information gain. Add what the cited competitor pages lack: your data, your process, your screenshots.
  6. Measurement. Re-inspect in two weeks and compare AI referral sessions month over month.

Frequently asked questions

How can I tell if my site has these problems?

Open Search Console → Pages and read the “Why pages aren’t indexed” table, then check “Last read” under Sitemaps. If “Discovered” and “Crawled – currently not indexed” together cover a large share of your URLs, start with mistakes 1 and 3.

Will rewriting AI content to sound more human fix it?

No. Google’s policy targets pages that add no value, and it lists automated rewording as part of that pattern. Adding facts only you have changes the assessment; changing the wording does not.

How long does recovery take?

We don’t have a reliable number and won’t invent one. Recrawling alone takes days to weeks per URL. We are re-inspecting the pages above daily and will update this article with what we measure.

Not sure whether answer engines can understand, trust, and cite your business? Ask Lexington Digital about an AI Search Readiness Audit.