Loudmink has tracked 25 B2B SaaS brands across five AI search engines every week since spring 2026, and engine behavior keeps moving in both directions. Perplexity ignored Reddit for months, then began citing it in roughly 85 to 90% of its answers and held there for three straight weeks. Claude, long the most startup-friendly engine, saw its startup citation rate fall from a peak near 44% to about 18% in June. Set-and-forget AEO fails because the engine you optimized for last month may behave differently this month, and this article shows what shifted, why it matters, and how to monitor each engine continuously.
These are not one-week blips or noise. They are directional changes in how individual engines source their answers, and they reward the marketers who keep watching over the ones who optimize once and walk away.
The Bottom Line
- Perplexity flipped on Reddit. It went from citing Reddit zero times to citing it in most of its answers, turning Reddit from a one-engine behavior into a two-engine one.
- Claude is sliding on startups. Across the full quarter Claude still averages the highest startup citation rate, but its June direction is down and it is no longer the runaway favorite.
- Citations do not sit still. Only about 1 in 10 specific citations survived the entire quarter, so a visibility win this month is not a win you can bank.
Perplexity started citing Reddit after ignoring it for months
For its first stretch of weekly cycles, Perplexity cited Reddit zero times, then in the back half of June it began citing Reddit in roughly 85 to 90% of its answers and held there for three straight weeks. That makes Reddit a two-engine behavior now, not the Grok-only quirk it used to be. Grok still leads by raw volume, linking to about five times as many Reddit pages as Perplexity in June, but it is no longer alone. Claude still never cites Reddit at all.
This extends an earlier finding rather than contradicting it. When we reported that Perplexity began citing Reddit after never touching it, the open question was whether the shift would stick or revert. It stuck. The "Perplexity never cites Reddit" framing was accurate when it was published, and the engine simply changed.
What to do: If you wrote Reddit off because "only Grok uses it," reassess. Reddit now feeds two of the five engines we track, so a presence in the threads your buyers read is worth more than it was a quarter ago. Treat any single engine's current habit as a snapshot, not a permanent rule.
Claude is still the average leader on startups, but it is sliding
Claude recommended startups as often as about 44% of the time at its late-spring peak, and in June that fell to roughly 18%, dropping it below Gemini, ChatGPT, and Grok on startup citations. Over the full quarter Claude still averages the highest startup rate at about 24%, so calling it the most startup-friendly engine is still defensible on average. The direction, though, is clearly down, and the runaway lead is gone.
As of June 2026, here is how the five engines rank on startup citations:
| Engine | Startup citation rate in June |
|---|---|
| Gemini | ~23% |
| ChatGPT | ~22% |
| Grok | ~19% |
| Claude | ~18% (down from a ~44% peak) |
| Perplexity | ~5 to 6% |
Perplexity remains the toughest engine for young companies, sitting effectively near zero. We documented that gap when Perplexity posted a 0% startup citation rate against Claude's 44%. Both halves of that comparison are now moving: Claude is coming down toward the pack while Perplexity stays low. The 44% figure was a real peak, not an error, which is exactly the point. Peaks reverse.
What to do: If your startup won Claude citations in spring, do not assume they hold. Diversify your third-party presence so your visibility does not depend on one engine's current mood, and check Claude specifically each month rather than trusting last quarter's standing.
This is directional change, not run-to-run randomness
Month-over-month shifts in engine behavior are a different problem from the variation you see when you ask the same question twice in one afternoon. Run-to-run randomness is the built-in noise covered in why AI search results change every time you ask: the same prompt returns slightly different answers because the engines add randomness by design. That is short-term jitter around a stable average.
What we are describing here is the average itself moving. Perplexity citing Reddit in nine of ten answers for three weeks running is not jitter, it is a new baseline. Claude halving its startup rate over a month is not a sampling artifact, it is a trend. The two problems compound: you have to see through the daily noise to detect the slower drift underneath it, which is hard to do from a single manual check.
What to do: Measure the same queries on a fixed cadence so you can separate the daily wobble from the monthly direction. One reading tells you almost nothing. A line over several weeks tells you whether an engine is drifting toward you or away from you.
Only about 1 in 10 citations survived the whole quarter
Roughly 9% of the specific citations we tracked were present in every single weekly cycle across the quarter, which is the clearest evidence that the landscape is not settling. Even the category leaders moved: the most-cited brand of the quarter still swung between three and ten citations per cycle, and several well-known analytics brands slid or dropped to zero in the final June cycle. We cover this in depth in our analysis of how quickly AI citations decay, and it is the metric form of everything above.
The practical reading is simple. If only one in ten citations holds for a quarter, then any audit that captures a single moment is mostly capturing things that will not last. Comparison and head-to-head content produced the only citations that stayed rock solid all quarter, which is a useful tell about what to build.
What to do: Own the durable query types. Publish comparison and alternatives content on your own domain, keep it updated monthly to stay in the freshness window AI search engines favor, and track your citations over time so you notice the day one of them quietly disappears.
What to do when AI search engines keep changing
The fix for a moving target is continuous measurement, not a one-time optimization. Each of the five engines moves independently and in different directions, so a quarterly snapshot or a single audit will miss the shifts that matter. The four habits below cover most of it.
- Do not optimize for one engine's current quirk. Perplexity ignoring Reddit was treated as a fixed fact for months, then reversed inside a single month. Build for the behavior that is durable, not the habit that happens to be true today.
- Re-check monthly, per engine. The five engines do not move together. Watch each one separately so a slide in one does not hide behind strength in another.
- Reassess Reddit. It now matters for two engines, not one. A Reddit footprint that looked optional a quarter ago is closer to table stakes.
- Diversify and monitor. A win in one engine can reverse, so spread your presence across multiple third-party sources and keep watching so reversals surface while you can still react.
Loudmink is an AEO platform that monitors all five AI search engines separately on 24-hour cycles, so directional shifts like these surface as they happen instead of at the next quarterly review. Plans from $99/mo.
Frequently Asked Questions
Why does my AI search visibility keep changing month to month?
Because the engines themselves change how they source answers. Over a single quarter we watched Perplexity go from never citing Reddit to citing it in most answers, and Claude cut its startup citation rate by more than half. These are directional shifts in engine behavior, not just the run-to-run randomness you get from asking the same question twice.
Did Perplexity really start citing Reddit?
Yes. After citing Reddit zero times for months, Perplexity began citing it in roughly 85 to 90% of its answers in the back half of June 2026 and held there for three straight weeks. Reddit citation is now a two-engine behavior shared by Grok and Perplexity, though Grok still links to about five times as many Reddit pages.
Is Claude still the best AI search engine for startups?
On a full-quarter average, Claude still cites startups the most, at about 24%. But its June rate fell to roughly 18%, below Gemini, ChatGPT, and Grok. So Claude is the average leader that is trending down, not the runaway favorite it was in late spring.
How often should I check my AI search visibility?
At least monthly, and per engine. A single check captures a moment that may not last, since only about 1 in 10 citations survived our full quarter. Tracking the same queries on a fixed cadence is the only way to separate daily noise from the slower month-over-month drift.