Why is my site not cited by AI when the technical SEO is perfect?

Usually because it never ranks high enough to be read. Assistants that answer with live search build their response from a handful of top results, so a flawless page at position thirty is never among the pages the model reads. In the audit behind this entry every crawler got full HTML and every technical check passed, yet the pages sat between positions 20 and 60 with almost no independent links pointing at them.

Can structured data get my site cited by AI?

It helps identification, not selection. Schema tells machines what a page is and who is behind it, but it cannot add the independent mentions and links that decide whether the page ranks high enough to be retrieved. A site with six valid schema types and no authority stays uncited, because markup is never seen on a page that is never pulled into the answer.

What should I do after a clean technical SEO audit?

Stop auditing and spend the time off site. Once crawlers get full HTML, canonicals are correct and structured data validates, further code changes return very little; the work that moves rankings is being mentioned in the roundups, alternatives lists and community threads that already rank for your category. Keep automatic indexing pings in place, but expect them to speed discovery rather than change position.

Does IndexNow improve rankings or AI citations?

It speeds discovery, not ranking. In the audit behind this entry, adding an automatic IndexNow ping got all 135 pages accepted for submission and new pages seen sooner, and it moved no rankings at all. That is the expected result: IndexNow tells engines a page changed, while whether the page is chosen still depends on authority and relevance.

Perfect Technical SEO and Still Not Cited

¶ WAR STORYTHE AUDIT THAT FOUND NOTHING

01Every box ticked, and still absent from every answer

At the end of July I audited a free utility site I run that was not getting cited anywhere. I crawled it the way the machines actually see it, with JavaScript switched off and the real user agents of every crawler that matters, expecting to find the bug within an hour. I was sure there had to be one. A site that useful, that invisible, had to be broken somewhere.

What passed

Everything passed. Every AI crawler and both big search crawlers got a 200 and the full page in plain HTML. The sitemap listed 135 addresses with consistent formatting and dates. Every page I sampled pointed its canonical at itself, allowed indexing and carried between 670 and 3,700 words of original text. Six types of structured data, all valid. Around 25 contextual internal links on each landing page. Instruction files for AI agents, present and accurate. An audit can prove you are reachable. It cannot prove you are wanted.

// THE AUDIT THAT PASSED~~A FREE UTILITY SITE · JULY 2026~~

✓Every AI crawler and both search crawlersHTTP 200, full HTML without JavaScript

✓robots.txt and sitemapClean, 135 URLs, dates present

✓Canonicals and indexingSelf referencing, index and follow

✓Content depth670 to 3,700 words of original text a page

✓Structured dataSix types, all valid

✓Internal linkingAbout 25 contextual links a landing page

✓AI instruction filesBoth present and accurate

✕Independent sites mentioning itAlmost none, against category leaders with years of links

Every row a crawler can test passed. The failing row is the one no crawler can fix.

¶ WAR STORYWHERE IT ACTUALLY LOST

02Retrieval is a shortlist, and the site was never on it

The explanation was sitting in Search Console the whole time. In July the site collected roughly 8,000 impressions and 24 clicks, a click rate near 0.3%, with most of its pages ranking somewhere between position 20 and position 60. Nothing was blocking the pages. They were simply far down the pile.

Retrieval takes the top of the pile

An assistant answering with live search does not read everything that matches. It takes a handful of results near the top and builds its answer from those. A perfectly marked up page at position thirty is still at position thirty, which means it is never among the pages the model reads before it writes, so none of that careful markup is ever seen. You cannot schema your way out of being unknown.

Authority was the missing row

The pages ranking above it belonged to the established names in its category, companies with years of links from sites that had written about them. This site had almost none. No change to templates, schema or speed adds an independent mention, which is the thing that was missing, and structured data was never going to supply it, because schema makes you identifiable, not interesting. Technical SEO decides whether you can be found. Authority decides whether you are chosen.

Authority in this sense is less mysterious than the word suggests. It is the accumulated evidence, mostly links and mentions from sites that had no reason to write about you, that a page deserves to sit near the top. Search engines use that evidence to order their results, and every assistant that searches before answering inherits the order the moment it runs the query.

EXTRACTED — THE SENTENCE THIS ENTRY EXISTS FOR

The crawl came back clean. That was the bad news.

¶ WAR STORYWHAT ACTUALLY HELPS

03The fix is not in the repository

This blog got the same diagnosis

A week later I ran the same kind of audit on this blog, which had zero impressions in both Google and Bing. Structure was fine: valid canonicals, question and answer markup, more than a thousand words in the raw HTML, crawlers visiting. The real causes were plainer. The section was twelve days old, it had roughly one backlink, it targeted questions almost nobody was typing yet, and Bing's crawler had visited four times in its life. None of that shows up as an error, which is the lesson of the reports that scream and the defects that stay silent. A clean audit is a reason to stop auditing, not a reason to relax.

Spend the next month off site

Once a crawl comes back clean, every further hour spent on the codebase has a falling return. The work that moves a site from position thirty into the pages a model reads is being mentioned by others: in the roundups that already rank for your category, the alternatives lists, the community threads where people ask what to use. That is the same arithmetic as the pages you do not own, arriving from a different direction. When the crawl is clean, the next hour belongs on somebody else's page.

In practice that means starting with the lists that already rank for your category's main question, asking whether each has a place you could reasonably earn, and making the facts an editor would need easy to copy. A month later, count how many independent pages name you. That count, not your audit score, is the number that predicts whether an assistant ever reads you.

The audit did find exactly one real gap. The site was not announcing new pages to search engines, so I added an automatic ping and the first submission came back accepted for all 135 addresses. It helped new pages get seen faster and it did not move a single ranking, which is precisely what it should do. Being seen sooner and being chosen are different problems.

Tomorrow: what an assistant says when you ask it what happened to your company.