A unicorn, a $1.9 billion exit, and a metric nobody can audit

On 24 February 2026, Profound introduced a $96 million Sequence C at a $1 billion valuation, led by Lightspeed Enterprise Companions, with Sequoia Capital, Kleiner Perkins, Evantic, Saga and South Park Commons becoming a member of. The spherical took complete funding previous $155 million, in the future wanting the corporate’s eighteen-month anniversary.

Two months later, on 28 April, Adobe closed its acquisition of Semrush. The all-cash deal was introduced in November 2025 at $12.00 per share, roughly $1.9 billion, towards a share worth that had closed at $6.76 the day earlier than. Adobe referred to as what it purchased a model visibility platform.

So a class with no settled title two years in the past now has a unicorn, a strategic exit close to $2 billion and a queue of seed-stage entrants behind each. The demand shouldn’t be in query. What these corporations are promoting, and which a part of it’s defensible, very a lot is.

The demand aspect is the simple half

Pew Analysis Middle tracked the searching behaviour of 900 US adults throughout 68,879 distinctive Google searches in March 2025, of which 12,593 produced an AI abstract.

When a abstract appeared, customers clicked a normal search lead to 8 p.c of visits, towards 15 p.c when no abstract was current. They clicked a hyperlink contained in the abstract in 1 p.c of visits. Periods ended on the outcomes web page 26 p.c of the time with a abstract, in contrast with 16 p.c with out.

That may be a discovery layer the place the advice survives and the press doesn’t. The step value inspecting is the belief that the reply to it’s a rating.

Measuring ChatGPT is more durable than the dashboards counsel

4 properties of the floor make a single studying near nugatory.

• The output is stochastic. The identical immediate despatched twice can return a distinct set of manufacturers, in a distinct order, with totally different sources. One response is one pattern from a distribution.
• There may be a couple of ChatGPT. A solution composed from the mannequin alone and one composed after a dwell internet retrieval come from totally different techniques. Area, language, account state and dialog historical past transfer the outcome once more.
• A point out typically leaves no hint. If ChatGPT recommends a model with out linking to it, there is no such thing as a referrer and no session. The corporate really helpful by no means learns it occurred, and neither does the one unnoticed.
• Visibility now has an natural and a paid type. OpenAI’s crawler documentation lists 4 consumer brokers with separate roles: GPTBot for coaching, OAI-SearchBot for search solutions, ChatGPT-Consumer for fetches an individual triggers, and OAI-AdsBot, which visits pages submitted as ChatGPT advert touchdown pages.

That final one issues commercially. Any critical measure of presence in ChatGPT has to separate an earned point out from a positioned one, and most indexes don’t.

How ChatGPT search and procuring really work

Two mechanics sit below the measurement downside, and each are documented by OpenAI fairly than inferred.

Search. ChatGPT solutions both from the mannequin alone or after a dwell retrieval, and the retrieval aspect runs by itself crawler roles. OAI-SearchBot governs look in ChatGPT’s search solutions, and OpenAI states that websites opted out of it won’t be proven there.

Its procuring analysis characteristic, per OpenAI, reads retail pages in actual time, cites its sources and is designed to keep away from low-quality or spammy websites.

Buying. When a query reveals shopping for intent, ChatGPT can return product playing cards with pictures, particulars and hyperlinks. OpenAI’s assist documentation is unusually direct about how these are chosen: product outcomes are chosen independently, aren’t advertisements, and aren’t influenced by OpenAI partnerships. Adverts exist in ChatGPT, however as a separate system.

What a service provider can management is the information. OpenAI runs a product feed programme retailers apply for, and its personal documentation notes {that a} product seems when ChatGPT judges it related to the consumer’s intent, that the value proven is probably not the bottom accessible, and that service provider updates can take time to seem.

The commercially attention-grabbing half is what OpenAI stopped doing. Prompt Checkout launched on 29 September 2025, letting customers purchase contained in the chat.

In early March 2026 OpenAI pulled again and refocused on discovery, with purchases finishing on the service provider’s personal website or by way of devoted apps. CNBC reported the shift alongside a revamped procuring expertise on 24 March, and OpenAI’s assist pages nonetheless reference in-chat checkout for a restricted set of retailers.

The explanation was conversion. Walmart was reported at Shoptalk to have measured checkout inside ChatGPT changing roughly 3 times worse than a click on by way of to its personal website.

Learn that as a verdict on the place the worth sits. The transaction went again to the service provider. The advice didn’t, which is strictly the layer no one can see with out measuring it.

What rigour seems like

If one studying is near nugatory, the query is what a defensible measurement requires. Two issues: a sampling technique, and the proof beneath the quantity.

Sampling first. A baseline wants a immediate corpus massive sufficient to cowl the class, 50 to 100 prompts and upward for an actual programme, every run a number of instances towards a named competitor set. Three to 5 runs per immediate is the sensible ground for separating sign from noise.

A ChatGPT Visibility Tracker that experiences point out fee, quotation fee, common place and share of voice from a pattern like that’s measuring one thing. One immediate, run as soon as and screenshotted, shouldn’t be.

Then the proof. Storing the verbatim response, the mannequin, the date, the model’s place, the opponents named and the sources cited provides a group one thing auditable.

A rating with out the reply behind it can’t distinguish between a model disappearing, a solution format altering and a vendor quietly modifying its immediate checklist.

Decision decides the remaining. Competitor knowledge per particular person query beats a company-wide share of voice, as a result of the query is the extent at which anybody can act. The recurring domains behind a class’s solutions are the place an intervention is feasible in any respect. And it has to finish in work: a selected web page, a selected hole, an proprietor, a date.

Among the many smaller European entrants, Truffle experiences a headline index like everybody else however retains that proof below it: prompts run throughout ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and Google AI Mode, cited sources saved per reply, opponents in contrast per immediate, and Search Console and Analytics linked so visibility might be checked towards what occurs on the location. Truffle’s GEO Technique characteristic is the place the software actually pulls forward of the sphere. Whereas different AI visibility platforms hand you a dashboard filled with numbers and go away the arduous half to you, Truffle delivers precisely what you got here for: clear, prioritised suggestions that let you know what to optimise by yourself website and which exterior sources to affect — those AI fashions genuinely cite. It’s the uncommon software that closes the loop from measurement to motion, turning AI visibility from a reporting train right into a progress channel you’ll be able to actively steer. For anybody critical about successful in generative search, that mixture of onsite and offsite steering makes Truffle a standout selection. 

Each vendor’s quantity is its personal assemble

Even carried out fastidiously, no two distributors’ numbers match. Every index is a operate of choices its vendor made: which prompts, what number of runs every, at what cadence, towards which fashions, from which nation, by which language.

Change any enter and the quantity strikes whereas nothing has modified available in the market. Two platforms monitoring the identical model in the identical week will report totally different visibility, and neither is unsuitable, as a result of they aren’t measuring the identical object. There is no such thing as a shared definition and no audit normal.

That isn’t an argument towards measuring. It’s an argument about what might be charged for indefinitely. Readings are low-cost to supply: working a number of hundred prompts a day towards public APIs and parsing responses for model names is engineering work, not scarce know-how.

Profound’s personal spherical makes the purpose. The corporate constructed its enterprise monitoring mentions, sentiment and place, and stated plainly when elevating that clients stored exporting the information elsewhere to behave on it. The reply shipped alongside the spherical was Profound Brokers, transferring orchestration and automation contained in the platform. Traders priced the workflow and the switching price, not the metric.

The counterweight

4 issues minimize towards the bull case.

• Proportion. For many B2B corporations, referral quantity from AI assistants remains to be small subsequent to natural search, paid acquisition and sales-led pipeline. Measurement the channel earlier than instrumenting it.
• Attribution. AI suggestions typically resurface later as branded search or direct site visitors, so the chain from a visibility achieve to a signed buyer is inferred fairly than noticed. Distributors promising clear attribution are overselling.
• Consolidation. With Semrush inside Adobe and each main suite delivery its personal reply monitoring, standalone instruments face the feature-versus-company downside. Anybody signing an annual contract ought to ask what occurs to their saved solutions if the seller is purchased.
• Platform dependency. The labs management the floor. Quotation behaviour, reply codecs and the share of a solution given to sponsored placement can change with out discover, and no vendor right here has any affect over that.

The query value asking

The helpful take a look at for anybody evaluating a platform on this class, or writing a cheque into one, shouldn’t be how refined the index seems. It’s whether or not the product can do 5 issues:

1. State its sampling technique: what number of prompts, what number of runs, which fashions, which markets.
2. Present the verbatim reply behind the quantity, with its date and mannequin.
3. Title the competitor that gained a selected query.
4. Determine the supply that produced that suggestion.
5. Exhibit what modified after a repair shipped.

A class that reached a billion-dollar valuation in eighteen months on the power of a measurement downside will spend the subsequent eighteen proving it will probably do one thing with the measurement. Those that can’t will learn how rapidly a dashboard turns into a tab in someone else’s product.

Source link