Two structural things happened to organic and AI search in the same ~30-day window: the Google March 2026 core update finished rolling out April 8, and OpenAI shipped GPT-5.3 Instant as the default ChatGPT model in late January with downstream effects landing through Q1.
One reshuffled traditional rankings; the other rewired how AI engines hand off traffic to source sites. We’re observing most marketers reading the chart and reaching the wrong conclusion: “AI search is collapsing.” It is not. (I repeat: It is not.).
Here’s what we’re seeing across our portfolio of B2B SaaS companies and how we’d think about it.
The piece to read
If you missed last week’s Field Notes, stop now, open a new tab, and read Ligia’s latest piece: Cited, not clicked: The new math of AI traffic.
TL;DR (yes, people still use this): AI-driven referral traffic across our book started bending downward in late January and worsened through March, but visibility scores were flat or up across the same window. AI engines were citing our clients more, not less.
The drop was concentrated in one platform (ChatGPT), tied directly to GPT-5.3 Instant compressing citations from 10-12 sources per answer down to 2-3 max. The “missing” sessions are mostly showing up as Direct/None in GA4 because mobile in-app browsers strip the referrer header.
A Wheelhouse DMG analysis estimates GA4 may be capturing as little as 9% of true AI-referred mobile traffic.
Ligia’s piece runs through our team’s diagnosis, the platform-level data, the BOFU-content angle (where citation compression hurts most), and how we’re shifting our measurement framework.
If you’ve been watching your ChatGPT referral line slide and worrying about your content, read this.
What we saw across the portfolio in April
A few patterns held across most of our active accounts.
The median client saw organic sessions drop about 14% MoM, with the spread running from -45% to +24% and most clients clustering between -14% and -22%. Two clients posted positive MoM organic in April: a security-awareness platform was up +24% (top-decile for the month, driven by BOFU competitor-comparison content that institutional publishers don’t necessarily compete on), and a Canadian fintech client was up +1.6% to a new high.
About 80% of top-3 results shifted on tracked queries during the core update, the most volatile reshuffle since December 2025.
Of our clients with LLM-session tracking in Profound or Peec.ai, ~90% saw LLM-referred sessions decline in April, with a median drop of 17%. The single outlier grew LLM sessions 14% MoM, attributed to our Reddit engagement program running in parallel to SEO/GEO work.
So the redistribution we wrote about above is real: ChatGPT referrals are down, Gemini and Perplexity are flat or up.
The conversion-quality story is the more important one. Across the same set of clients, conversion efficiency from LLM-referred users improved MoM even where session volume dropped. ~40% of clients tracking LLM-attributed conversions posted positive MoM growth in April even where sessions fell.
- One hosting client hit a record 1.79% LLM-sessions-to-lead conversion rate; AI-sourced leads grew 14% MoM despite flat AI session volume.
- A board-management client saw LLM-attributed conversions grow 33% MoM, the highest LLM-conversion month for that account since November.
- The Canadian fintech I mentioned above put up its strongest conversion performance in our entire tracked period: organic conversions +20% MoM on a +1.6% session lift, and LLM conversions +18% MoM even as LLM sessions dropped 17%.
The takeaway? Don’t let the traffic loss spook you if your conversion line is going up and to the right. Call it Traffic Trap 2.0—don’t fall for it!.
Three things working right now
Refreshing existing high-intent pages is outperforming new content for the third month running.
AI Overviews are eating the click on broad informational queries, but BOFU pages with intent-laden modifiers (best, alternatives, vs., for [vertical]) still convert and still rank.
One inventory software client we work with grew organic registrations +200% MoM in April, driven entirely by refreshes.
If your team has a backlog of pages that ranked top-3 in 2024, drove real business, and have since slipped, that backlog is the highest-leverage place to invest content effort right now.
Reddit isn’t a marketing channel; it’s a strategic brand play.
Reddit shows up as a primary source in AI-generated responses across virtually every B2B software category, which means the question for any brand whose buyers research on Reddit is not “should we be there” but “are we shaping how AI engines describe our category?”
We just published our learnings from building our Reddit engagement program and how we’ve iterated alongside changes to the platform and its influence on LLMs.
(Hint: Your brand account is the brand mention.)
Speaking of… brand mention outreach is a faster GEO lever than most teams realize.
When citation slots compress from 10 to 2, the model gravitates toward third-party authority signals: G2, Reddit, industry publications, original research, and named experts. The brands still showing up in those two slots are the ones who built those signals before the compression happened and continue to invest there. Our research supports this.
Three under-used citation surfaces worth auditing if you haven’t recently:
- LinkedIn Products pages (most B2B brands haven’t configured these and they’re cited often)
- Medium author pieces
- PartnerStack-backed listicles
These are warmer outreach paths than cold digital PR and they show up in citation tracking faster than backlinks.
On the radar
A few things we’ve started spotting in April:
.gov and other authoritative publishers are eating TOFU informational queries.
If your top-of-funnel content drove material traffic in 2024, expect compression on those terms.
The defensible response is to push content effort down-funnel toward comparison, alternatives, and product-adjacent pages where institutional publishers don’t compete.
UGC platforms are quietly climbing post-update.
Our team has been spotting Medium.com dominating several SERPs we track, and a quick Ahrefs check confirms Medium has been trending up since the March core update. Substack and Ghost are also gaining top-10 positions at a similar rate.
If you have a thought-leader on your team without a Medium presence, the cost to test this is low and the timing is right.
LinkedIn Products listings are starting to show up in BOFU SERPs.
As an example, the LinkedIn recruiting-software category page is now ranking on BOFU queries in that space.
Product listings are free to enable on LinkedIn company pages (the ordering logic is opaque, but presence is half the battle). If your category has a LinkedIn Products page and you haven’t configured your product on your company page, this is a quick task with potential SERP and citation upside.
Closing
If your dashboard says ChatGPT traffic is declining, the problem isn’t your content. The problem is that the dashboard is reading one number in isolation that no longer tells the full story on its own.
If that’s the kind of question you’ve been wrestling with internally and you’d like a second pair of eyes, book a strategy call and we’ll dig in.
Want more insights like this? Subscribe to Field Notes
