Engine log: 11 changes this week, one rollback, one surprise

Key takeaways

• 11 changes shipped in week one of August; one rollback (meta-description rewrites that tested well in click-through but reduced dwell time by 18%).
• Adding FAQ schema to eight existing posts produced measurable citation lifts within four days — faster than any new-content experiment this quarter.
• The surprise: two posts updated only with a dateModified bump and a single revised paragraph each earned new Perplexity citations within 72 hours.
• The rollback lesson: optimizing for CTR without measuring post-click behavior is how you improve a proxy while degrading the outcome it represents.

Week 1 of August. Eleven changes shipped across nine posts, one rollback, one result that changed how we're prioritizing the queue. This is the engine log format: what ran, what measured, what it means.

FAQ schema additions (8 posts)

We added FAQ schema blocks to eight existing posts — posts with established authority, existing rankings, and no structured Q&A. No other changes to those posts.

Citation tracking at three intervals:

• Day 1: No new AI citations across any of the eight posts.
• Day 4: Three of eight posts showed new AI citations — two in Perplexity, one in a ChatGPT Browse response.
• Day 7: Five of eight posts showing citations. Three of the five are citing the FAQ answer text directly, not the surrounding body copy.

Why FAQ schema works at this speed: answer engines need a pre-chunked, question-shaped unit of content. A paragraph that contains an answer is harder to extract than a block explicitly typed as question + answer. FAQ schema does the chunking work for the crawler. The engine doesn't have to infer where the answer starts and stops.

The speed advantage over new-post creation is real. These posts already had domain authority and topical relevance. Adding FAQ schema gave the crawler a new extraction surface on content it already trusted. A new post on the same topic would need to earn that trust first — typically 3–6 weeks before meaningful citation activity.

dateModified bumps and paragraph revisions (2 posts)

Two posts received minimal edits: a dateModified timestamp update and a single revised paragraph each. The paragraph revisions tightened the lead sentence and made the core claim more extractable. No structural changes, no new sections.

Both posts earned new Perplexity citations within 72 hours of the update.

This is the clearest signal we've seen on the update-vs-create decision. AI crawlers treat freshness as a quality signal — a recently modified authoritative post gets re-evaluated sooner than a stale one. A targeted revision to a post that already ranks and already has topical authority can outperform a new post on a related topic, because the new post starts from zero on both dimensions.

The practical implication: before creating a new post on a topic, check whether an existing post covers 70% of it. If it does, revise and re-date. The marginal cost is lower and the citation timeline is faster.

Full body rewrite (1 post)

One post received a complete body rewrite — new structure, new examples, same topic and URL. The rewrite was driven by a content audit flag: the original post had high impressions but low dwell time, suggesting the content wasn't delivering on the title's promise.

Results pending. It's too early to measure citation impact from a full rewrite; we'll log this in week 3.

Structural changes

Two additional changes ran this week that don't fit the content categories above:

• Internal link additions: Four posts received new internal links pointing to a pillar page that had no inbound links from the blog. This is a PageRank distribution fix, not a content play. The pillar page now has link equity flowing to it.
• Schema additions beyond FAQ: Two posts received HowTo schema blocks where the body already contained numbered steps. No content was changed — the schema was added to match structure that already existed.

Neither of these is expected to move citation metrics directly. They're infrastructure.

The rollback: meta-description rewrites

This experiment ran for five days before we killed it.

We rewrote meta descriptions on six posts to be more keyword-dense — front-loading the focus keyword, tightening to under 155 characters, matching search intent more explicitly. CTR improved 11% over the five-day window.

We also measured dwell time and bounce rate. Dwell time dropped 18%. Bounce rate worsened.

We reverted all six.

The lesson is straightforward: CTR is a proxy for interest, not a proxy for satisfaction. A more compelling meta description gets more clicks from people who expect more than the page delivers. When the page doesn't match the promise, they leave faster. You've optimized the proxy while degrading the outcome the proxy is supposed to represent.

This is a common failure mode in SEO. The metric you can see (CTR) moves in the right direction. The metric you care about (did the visitor get what they came for) moves in the wrong direction. If you're not measuring both, you don't know which way you're actually going.

What changes going into week 2

The week's data shifts two things in the queue:

FAQ additions and targeted revisions move up. Both showed faster citation impact than new-post creation, at lower production cost. We're running a full audit of existing posts to identify FAQ candidates and revision targets.

New-post creation is not the default lever. It was. It isn't now. New posts are the right call when there's no existing coverage of a topic. When coverage exists, revision is faster and the citation timeline is shorter.

The rollback is a reminder to instrument the thing you care about, not just the thing that's easy to measure.

Frequently asked questions

What does Avakata's engine actually change week to week?

Changes fall into three categories. Content: FAQ additions, body rewrites, and dateModified updates on posts where freshness affects ranking. Structural: internal link additions and schema markup changes (FAQ, HowTo, Article). Experiments: tests that ran, produced a result, and were either kept or killed. Every change is tied to a measured outcome — organic impressions, click-through rate, dwell time, or AI citation frequency. Nothing ships on intuition alone.

When should you roll back a content change?

Roll back when a proxy metric improves while the outcome metric degrades. In the meta-description experiment, CTR went up after the rewrite — the proxy looked good. But dwell time dropped and bounce rate rose, meaning the new description was attracting clicks from users who left immediately. The change was optimizing for the wrong signal. When a proxy and the thing it's supposed to represent diverge, the proxy is lying. Revert and reframe the test.

Why do FAQ schema additions lift AI citations faster than new posts?

FAQ schema gives AI answer engines pre-chunked, question-shaped content units that are trivial to cite verbatim. A new post has to earn domain authority from scratch. An existing post already has it — backlinks, crawl history, engagement signals. Adding FAQ schema to that post surfaces existing authority in a format optimized for AI retrieval. The engine isn't building something new; it's making what's already there easier for AI systems to extract and quote.