AEO
ChatGPT's citation behavior after the Q1 2026 model swap, what to fix this week
The Q1 model update reweighted how ChatGPT picks B2B citations. Here's what we observed across 600 buyer-style queries, and the three things every AEO program should refresh now.
What changed
The Q1 ChatGPT update, rolled out in waves between mid-February and late March 2026, shifted three things in how it picks citations for B2B queries:
- Recency weighting jumped. Sources older than 12 months are ~40% less likely to appear in a buyer-style answer compared to the previous behavior. Evergreen content still gets cited; stale content gets quietly dropped.
- Self-cited vendor pages lost weight. Marketing-page content on the vendor’s own domain now needs to be paired with at least one third-party corroboration to surface. Pages that were citation anchors before the update have stopped appearing.
- Reddit citations rose. Subreddit threads, particularly older, high-upvote ones, appear in roughly 18% of B2B answers we sampled, up from ~11% in Q4 2025.
We ran 600 queries across 14 B2B categories in the second week of April to baseline the new behavior. The patterns are consistent across categories.
What to fix this week
Fix 1: Run a freshness pass on your top 10 pages.
Pull the pages that account for 80% of your AEO citations. Refresh the dates, update at least one statistic on each, and ping the search consoles. This isn’t optional anymore, it’s table stakes.
Fix 2: Stop citing yourself.
Audit your top 5 AEO pages. If the claim relies on “we” or “our” as the only source, find a third party that says the same thing and add the link. Customer reviews, analyst reports, independent benchmarks, any earned source that breaks the self-cite loop.
Fix 3: Find the three Reddit threads your buyers actually read.
Search Reddit for “$category recommendations” or “$category alternatives.” Filter to threads with 50+ comments, sort by recency. Read all three. If your product is missing or misrepresented, that’s the conversation to enter, named, in good faith, with permission from a senior person at your company. No sock puppets, ever.
The cadence going forward
Treat the AEO surface like an editorial calendar. Every quarter:
- 4 pages updated for freshness (no body rewrites, just date, stat, and 1-2 paragraph deltas)
- 1 new third-party corroboration secured for your top 5 claims
- 1 Reddit thread engaged on a buyer-relevant question
That’s enough motion to compound. More than that and you’re chasing the platform; less and you’ll fall out of citation reach within a quarter.
What we’re watching next
Two things on the horizon:
- Claude’s citation behavior is starting to diverge from ChatGPT’s in noticeable ways, it’s pulling more from documentation sources and primary research, less from review sites. If Claude’s market share keeps growing, the cite-graph optimization split widens.
- Perplexity’s source-link prominence has made it the click-throughest LLM in our analytics, buyers who click from Perplexity have ~3× the demo-request conversion of buyers who click from Google for the same query. If you’re not measuring Perplexity-sourced traffic separately, you’re missing the highest-quality referrer in your stack.
Former search-quality engineer. Reverse-engineers what makes LLMs cite, click, and convert, the cite-graph is the product.
Related
Playbook
Buyer queries you should run every Monday, a 15-prompt sanity check
Most teams run AEO audits quarterly and miss the weekly drift. Here's a 15-prompt Monday-morning routine that surfaces problems in 20 minutes a week.
Strategy
Stop optimizing for clicks. Start optimizing for citations
Click-through rate is a 2014 metric. In 2026, the LLM citation IS the impression, and the impression that compounds. Here's why your KPIs need to change, and what to measure instead.