A $30 Billion Perplexity Valuation, Built on Shaky Citations

The perplexity valuation pricing in a very young business#

Nvidia is reportedly in talks to invest in Perplexity at a valuation above $30 billion, according to SiliconANGLE, reported on 24 August, up from the roughly $20 billion valuation Perplexity finalised about a year earlier. No deal is confirmed and no size has been disclosed, but the number alone reframes what this company is now assumed to be worth.

The perplexity ai valuation only makes sense against the revenue climbing underneath it. Perplexity's own annualised revenue run rate has tripled this year, from under $250 million to $750 million, driven partly by Perplexity Computer, a cloud-based agent for professional tasks launched in April 2026. Whatever the final number Nvidia agrees to, a perplexity valuation above $30 billion assumes retrieval behaviour that markets can trust and plan around. That assumption had a very bad four days this August, which the rest of this piece treats as a case study, not a footnote.

Why Nvidia specifically#

Nvidia's reported interest reads as a bet on protecting its own chip dominance as much as a bet on Perplexity's product, positioning its silicon at the centre of whichever answer engine ends up owning the largest share of ai search market share once the current volatility settles. Perplexity has also previously taken money from Amazon founder Jeff Bezos and SoftBank Group, and the company is reportedly working toward a 2028 IPO, which gives Nvidia a stated timeline to realise a return rather than an open-ended bet.

Money arriving at this pace changes the stakes of a single bad week. A platform this well-funded is no longer a scrappy experiment nobody notices when it changes how it retrieves and cites. It is infrastructure that publishers, advertisers and now investors are pricing as though its behaviour were stable. The next section is the four days in August that tested that assumption directly.

Exhibit A: a citation share that vanished in four days#

A fox does not need a whole warren to vanish before it knows something has changed underground. One entrance thick with scent one evening and cold the next tells the same story. That, roughly, is what happened to Reddit inside ChatGPT Search this August, and it is the clearest evidence available for what a perplexity valuation is actually betting will not happen to Perplexity itself. A perplexity valuation this size is, in effect, a bet against exactly this kind of week. The site did not move house. Google could still find every thread. But the model that had been citing Reddit constantly simply stopped, and it stopped within days.

The tracking firm Promptwatch measured it in cold numbers, drawn from millions of live AI responses. Reddit held a steady 3.83% average share of ChatGPT Search citations from 18 July to 7 August. Between 14 and 17 August that average had fallen to 0.52%, an 86.4% relative collapse, first reported by Search Engine Land on 19 August. No ranking dropped. No policy changed in public. There was no algorithm update to blame, no manual action notice, no warning email.

Reading the trail back to 8 August#

The first tremor came six days earlier and it is where most of the honest analysis actually starts. Search Engine Journal's own reading of the Promptwatch dataset shows ChatGPT's use of site-scoped queries, the kind that target one domain directly rather than searching the open web, jumping 46-fold in a single day: from 0.37% of sampled queries to 16.8%. Average query fan-out, the count of related searches ChatGPT runs per prompt before it answers, nearly doubled at the same time, from 1.08 to 1.83. Then, six days later, for reasons still not fully mapped, the floor gave way completely.

Promptwatch is more careful than the headlines that followed about what its own 26-billion-point dataset can and cannot prove. Its published note calls the finding provisional and says the data shows when the shift happened, not why. That is the least comfortable part of this story for anyone pricing an ai search market share bet on the assumption that a platform's citation behaviour moves slowly, and with warning given first.

Why Google did not follow Reddit off the same cliff#

Here is the detail that matters most for anyone pricing risk into a perplexity valuation rather than just reading a headline: Google's AI features did not do this. Not at the same speed, not at the same scale, not even close, and that gap between two platforms sharing a story is the actual finding worth sitting with.

Plot each surface's Reddit citation share before 8 August against where it stood after, and the picture does the work no single headline number can. Google AI Overviews barely drifted, down from 2.37% to 2.10%, an 11% dip spread across weeks rather than days. Google AI Mode moved a little further, down from 2.22% to 1.54%, still nowhere near a cliff edge. ChatGPT Search sits well off that line, the one surface where the same publisher's fortunes changed by an order of magnitude while its two nearest cousins barely blinked.

Same word, different animal#

The irony worth noticing is that both companies use the identical phrase for the mechanism at the centre of this. Google's own documentation on AI features in Search describes AI Overviews and AI Mode running a 'query fan-out': a set of related searches fired behind the scenes before the summary is written. OpenAI's own web search tool documentation confirms ChatGPT's retrieval works the same way in principle, sending a query out, retrieving results, then choosing which sources earn an inline citation from the wider list it consulted.

Same term, two different retrieval engines, two entirely different outcomes for the same publisher in the same fortnight. A shared vocabulary does not mean a shared risk profile, and treating 'AI search' as one undifferentiated channel, with one undifferentiated ai search market share, is precisely the assumption this fortnight punished. It is also, by extension, the assumption underneath pricing any single platform's stock of goodwill as if it were fixed.

The market is already pricing citation risk, whether it says so or not#

Reddit's own week made the stakes personal rather than abstract, and it is worth reading alongside the same perplexity valuation talk rather than as a separate story. The company joined the S&P 500 on 18 August 2026, its shares having surged 11% on the announcement five days earlier, per KuCoin's market report. In that same week, as the citation charts began circulating publicly, RDDT stock dropped roughly 10% in a single afternoon session, according to Everything-PR's reporting on the episode.

A platform's citation share used to be a marketing curiosity, worth a slide in a quarterly deck. For a newly index-listed public company, it is starting to look like a number that can move a share price in an afternoon, and that is the exact category of risk sitting underneath any perplexity valuation built on a growth story rather than an audited, stable customer base.

Not the first time, and maybe not the last#

This is not without precedent, which is either reassuring or unsettling depending on how much trust you put in a pattern repeating. A near-identical collapse hit Reddit inside ChatGPT in September 2025, when its citation share fell from roughly 7% to roughly 1%, tracked separately by the AI visibility firm Profound across more than 4 billion citations, as recounted in Aivo's retrospective on the current episode.

That earlier collapse fully recovered within weeks, and its cause was later traced to a change in how data could be collected rather than a change in what ChatGPT valued. Almost nobody covering this month's version mentioned the earlier one. A rebound this time is not guaranteed, and treating last year's recovery as a promise rather than a single data point is exactly the kind of comfortable assumption a genuinely volatile channel goes on to punish, and exactly the kind of assumption a $30 billion figure needs to survive contact with.

What to actually do about a market you do not control#

None of this is a reason to abandon Reddit, ChatGPT, Perplexity or any other channel folkfox does not control, because nobody controls any of them entirely: that is the whole point of platform-mediated distribution. It is a reason to stop treating a citation count, or a headline perplexity valuation, as a fact you can bank and start treating it as a reading you take regularly, the way a fox checks the wind before it commits to a crossing.

The first move costs nothing and most brands still skip it: separate citation reporting from ranking reporting entirely. A page can hold position one in Google and still be invisible inside ChatGPT Search, and the reverse is just as common now. Keep them as two different numbers on two different lines, because conflating them is exactly what let this collapse pass unnoticed at most companies until their referral traffic did the explaining for them, weeks late and short on context.

Start with a real baseline, taken this week rather than remembered from a pitch deck. Google's own Search Console performance report still shows impressions, clicks and average position for classic organic results, and reading impressions against clicks tells you whether you are being summarised rather than demoted. That report says nothing about ChatGPT, Perplexity or Claude, which is precisely why it cannot be your only instrument.

None of this replaces the older discipline folkfox already builds for clients: self-contained, sourced sentences that survive being lifted out of context, entity structure that names who is speaking, and pages built to answer one query cleanly. What this month adds is a second habit sitting alongside it, measurement cadence, because a page, or a platform, can look strong on every visible metric and still be exposed to a retrieval change nobody announced. That work sits inside folkfox's SEO and GEO services, and it applies just as directly to regulated fintech brands as it does to a forum with 500 million weekly users.

Choosing an ai visibility tracking tool that tells the truth#

The unpredictability is not confined to Reddit, or to ChatGPT. A separate Ahrefs audit of 863,000 keywords and four million AI Overview URLs found that the correlation between holding a top-10 Google ranking and being cited inside an AI Overview fell from 76% to 38% in eight months, reported by Search Engine Journal on 2 March 2026. Ranking well used to be a reasonable proxy for being cited. It increasingly is not, on any platform, which is the actual argument for a monitoring habit rather than a quarterly report nobody rereads, or a valuation nobody stress-tests.

So pick a tool for what it samples and how often, not for its headline dashboard. A tool that pulls once a month would not have caught this collapse until it was three weeks old. A tool that pulls daily catches the shift inside a single cycle, the same way a fox catches a change in the wind before it reaches the clearing rather than after. The right choice depends on budget and the number of surfaces that matter to your business, but the cadence question comes first, always, before the vendor comparison.

The honest closing point is that no single number here, not Promptwatch's 0.52%, not Ahrefs' 38%, not folkfox's own reading, should be trusted in isolation, and the same discipline applies to any headline perplexity revenue figure, ai search market share estimate, or perplexity valuation you are handed in a pitch deck. Read our own investigation into Reddit's robots.txt timeline for the forensic detail behind this same collapse, and pair it with a live baseline of your own domain's citation share before you draw a single conclusion about what caused what. A measurement taken once, however alarming, is a photograph. Only a repeated one is a trend, and only a trend earns a strategy change, or a valuation.

Frequently asked questions#