What Is This Note?
This is a dated note on a change we made to our blog on 27 March 2026. It replaces a post called "How We Optimized 21 Posts for AI Citation in One Session". The changes it describes were real, and most of them are still on the site. The premise of the original post, that this formatting would get the posts cited by AI engines, is not supported, so this note records what we did and withdraws what we claimed.
What Did We Change?
The commit is c1f3df35, made on 27 March 2026. Its message lists the changes, and the table below follows it.
| Change | What it did | Still on the site? |
|---|---|---|
| Speakable markup | Added a SpeakableSpecification to every Article JSON-LD, pointing at the headline, the description and the TLDR | Yes |
| Entity-linked knowsAbout | Pointed the Organization and Person knowsAbout lists at knowledge graph entities by @id instead of plain strings | Yes |
| StatBlock component | A callout for one number with its source | Yes, with a named source now required |
| 24 StatBlocks in 21 posts | One to three numbers per post pulled into callouts | Some corrected or removed in October 2026 |
| TLDR data attribute | Let the speakable markup point at each post summary | Yes |
Which Claims From the Original Post Are Withdrawn?
The original post made claims about AI engines that we cannot support, and one about our own StatBlocks that turned out to be wrong.
| Claim in the March 2026 version | What the evidence shows | Status |
|---|---|---|
| The formatting makes posts citable by AI engines | Never measured; Google says (May 2026 guide) its AI features need no special formatting or schema | Withdrawn |
| A Princeton study proved a 30 to 40 percent visibility lift from statistics | A lab benchmark result; a July 2026 survey of 45 GEO studies found no stable cross-platform effect | Withdrawn |
| AI platforms scan top-down and pull tables almost verbatim | No source; an assumption about how engines work | Withdrawn |
| Every StatBlock has a verifiable, sourced number | Some carried unsourced numbers, including a $0.001 per-lead cost | Corrected |
| Results would follow in a later post | No follow-up was published | Withdrawn |
The source on Google is its guide to its generative AI features, published in May 2026. The source on the GEO research is a July 2026 survey of 45 GEO studies. The Princeton and Georgia Tech paper itself is a careful piece of work; the mistake was ours, in presenting a benchmark result as a general effect.
Why Do We Still Use These Components?
We still use StatBlocks, TLDRs and question headings because they make a post easier to read, scan and check. A number in a callout with its source beside it is easier for a reader to verify than one buried in a paragraph. That is an editorial reason, and it is enough.
The difference is in what we promise. Our blog standards now describe these patterns as making posts easier to scan, quote and verify, and say that they are not a citation lever. Our blog standards require a named source for every StatBlock.
What Changed in This Post?
We cut this post down to a dated note on 2 October 2026, as part of a review of every post on the blog. The changes:
- the premise that the formatting causes AI citations is withdrawn;
- the Princeton figure is no longer presented as a general effect;
- the "every StatBlock is sourced" statement is corrected;
- the step-by-step guide and the examples that relied on those claims are removed;
- the factual record of what changed on 27 March 2026 is kept.
For how we think about measuring AI visibility now, see Your SEO Is Fine. Your AI Visibility Needs Its Own Measurement.
The March 2026 Retrofit: Questions Readers Ask
Common questions about this topic, answered.
What did Pixelmojo change in 21 blog posts in March 2026?
On 27 March 2026, in commit c1f3df35, we added speakable markup to the Article structured data on every post, linked our Organization and Person markup to our knowledge graph by @id, created a StatBlock component and added 24 StatBlocks to 21 posts. The TLDR component gained a data attribute so the speakable markup could point at it.
Did the retrofit make AI engines cite those posts more?
We cannot show that it did. The original post promised to measure the effect and never published a result. Google says in its May 2026 guide that its AI features need no special formatting or schema, and a July 2026 survey of 45 generative engine optimization studies found no stable effect that holds across platforms. We now present the formatting as an editorial standard that makes posts easier to scan and check, not as a citation lever.
Did the Princeton GEO study prove that statistics raise AI visibility by 30 to 40 percent?
No. The Princeton and Georgia Tech study (KDD 2024) measured visibility changes in a lab benchmark it built for the purpose. The original version of this post presented that benchmark result as a general effect. A July 2026 survey of 45 GEO studies (arXiv 2607.14035) found no stable cross-platform effect, so we no longer cite the figure that way.
Were all 24 StatBlocks properly sourced?
No. The original post said every StatBlock carried a verifiable, sourced number. A review of the whole blog on 1 October 2026 found StatBlocks with numbers that had no source, including a $0.001 per-lead cost that was never measured. Those were corrected or removed in October 2026, and new StatBlocks need a named source.
What is speakable markup?
Speakable is a schema.org property, SpeakableSpecification, that points to the parts of a page suited to being read aloud. On our posts it points at the headline, the description and the TLDR. We still ship it, and we know of no evidence that AI engines use it to choose what to cite.
What does entity-linked knowsAbout mean?
Our Organization and Person markup list the topics we work on in a knowsAbout property. Since 27 March 2026, each topic points at an entity in our knowledge graph by @id instead of being a plain string, so a topic has one identifier everywhere it appears on the site.
What is a StatBlock?
A callout component for one number, with a label and a named source beside it. We added it on 27 March 2026 and placed 24 of them in 21 posts. Our blog standards now require a named source for every StatBlock.
Why keep these components if they are not a citation lever?
Because they make a post easier to read, scan and check. A number with its source beside it is easier for a reader to verify than one buried in a paragraph, and a short summary at the top helps a reader decide whether to keep going. That is an editorial reason, and it is enough.
