How to Check an Agentic AI Marketing Vendor's Claims

Every platform on your shortlist now says agentic. The claims you are comparing them on, the launch dates, the entry prices, the agent counts, the lift figures, almost all reach you second hand. A vendor's launch post becomes a round-up, the round-up becomes a directory row, and somewhere along that chain the qualifier that made the sentence true falls off.

So we took 20 claims about agentic marketing platforms out of our own competitive research and checked every one against the vendor's own primary source. Eight did not survive. One general availability date was out by six months. One entry price circulates at $1,500 a month and is actually $3,600. A bundle of agents quoted at $129 a month sits on a $249 tier. Ten of the twenty carried a figure a vendor had published about its own product, and not one of those came with a control group or a measurement method.

Every price, date and figure on this page was checked against the vendor's own pages on 8 September 2026.

None of that means the vendors are lying. Every one of those corrections came off the vendor's own site, which is the point. The market moved faster than the write-ups did, and a claim that was accurate in April is quietly wrong by September.

Why do agentic AI marketing claims fail when you check them?

Hardly anything in a vendor comparison comes from the vendor. The claim gets lifted from a launch post into a round-up, then into a directory, and each hop drops the qualifier that made it accurate: the tier it applied to, the date, the beta gate, the fact that the customer result belongs to the company selling the product.

The eight that failed were not exotic. Every one broke in an ordinary way, and each of those ways is something you can check in about two minutes on the vendor's own site.

Is the product actually shipped, and when?

An announcement date and an availability date are different facts, and they get flattened into one sentence constantly. Read the vendor's own press release rather than the coverage of it, then open the help centre page for the same product, because that is where the access gate gets written down.

Klaviyo's Customer Agent is the clearest miss. It circulates as a March 2026 general availability, and Klaviyo's own newsroom dates it to 25 September 2025, announced with Marketing Agent at K:BOS. Six months early. Klaviyo does carry a March 2026 release, on the 24th, expanding its AI agents, which is probably where the confusion starts. It added capability to an agent that had been generally available since the previous September, so it is not a general availability date. The nuance underneath matters too: WhatsApp is usually listed among the channels it launched with, and at general availability Klaviyo said chat, SMS, email, and soon WhatsApp. WhatsApp arrived later.

Braze is a case of three products collapsing into one date. The 23 April 2026 release does say BrazeAI Operator and BrazeAI Agent Console are now generally available. BrazeAI Decisioning Studio is not in that group. It had already gone generally available at Forge on 30 September 2025, and the April release only added European hosting on Google Cloud. The correct version makes Braze look further ahead on decisioning, not less, which is a decent sign you are reading the source rather than a competitor's summary of it.

Shopify is the reverse problem, where the round-ups describe a broader surface than Shopify's own documentation does. Sidekick is generally available. Campaign Autopilot, announced on 17 June 2026 to run marketing across Meta Ads, Microsoft Advertising, Shop Campaigns and Shopify Messaging, and Sidekick Pulse, which raises proactive recommendations in the admin, are both early access. Shopify's own help centre says each is available only to certain merchants, and Campaign Autopilot still sits behind a request form. Nor does it pause weak ads on its own judgement. Campaigns pause when they reach the budget you set.

Is the advertised price the price you will pay?

Entry prices are floors attached to a seat count, a contact tier and a feature set, and agent usage is increasingly metered on top of all three. Price the tier that actually contains the capability you came for, then add the onboarding fee and the usage meter before you compare anything.

HubSpot shows how far apart the two numbers sit. Marketing Hub Professional starts at $800 a month on an annual commitment, or $890 billed monthly, and that buys three core seats, 2,000 marketing contacts and 3,000 credits. Enterprise starts at $3,600 a month, not the $1,500 that circulates, which appears to be Content Hub Enterprise wearing the wrong label. Neither price includes the onboarding fee, charged once, at $3,000 on Professional and $7,000 on Enterprise. Breeze agents then bill separately in credits at $0.01 each, 50 credits every time HubSpot's Breeze Customer Agent closes a conversation without a human.

Okara sells an AI CMO, and round-ups routinely attach the full agent list to a $99 to $129 price. Okara's own pricing page does not describe it that way, and on 8 September 2026 it showed no $99 tier at all. The $129 Lite plan carries SEO, GEO, influencer, X and coding. The Reddit distribution, LinkedIn, content writer and UGC video agents are Pro only, at $249. The bundle in that quote costs $249, not $129.

Whose result is that number?

Ten of the twenty carried a figure a vendor had published about its own product. Six of those were quoted accurately and sit below. The other four failed on a separate point as well, so they are in the corrections table instead.

Five of the six say nothing about how the figure was measured. Attentive is the exception, and it is worth naming: it states that every AI Grow deployment is measured through incremental, cohort-level A/B testing against the brand's existing setup. The other five do not make that claim, which does not make their numbers false. It makes them unverifiable, which means they belong in your notes with the vendor's name welded to the sentence, never on a slide as a benchmark.

Two of these carry a second trap worth knowing about. The Cleo in Braze's story is the family care platform sold through employers, not the consumer money app of the same name, and readers attribute it to the wrong company constantly. And Attentive's numbers are medians across brands that opted into a test, which is a different population from every brand on the plan.

How old is the figure?

A number with no date attached gets read as current, and often is not. Before you trust a funding total or a price from a round-up, open the vendor's own about page and pricing page, because those move and the round-up does not.

Breinify is still listed at $11m across directories. That was the cumulative total in May 2021, after a $6m round. Breinify's own about page now shows $17m raised to date. The Blueshift figure is the more misleading of the two, because it points the wrong way. The $2,000 to $10,000 a month band traces to a single analyst post by David Raab in July 2017, and those were three separate per edition floors rather than a range, on a plan whose lowest tier started at 500,000 monthly active users. Blueshift's own pricing page today starts at $1,250 a month billed annually. Nine years on, the entry price came down.

Does the number measure what the sentence claims?

The quietest error is a real figure attached to the wrong noun. Read the words immediately around the number on the vendor's own page rather than the paraphrase, because the paraphrase is where the meaning shifts.

Ometria's 30 billion is real and sits on its own retail AI page, where it reads 30B retail customer data points used to train algorithms. That describes the training corpus behind Architect AI. It travels as personalisation delivered across 30 billion touchpoints, which is a delivery scale that page does not claim. Same number, a completely different assertion, and the second one is the sort of thing a vendor will ask you to correct.

What does a claim that survives look like?

Twelve of the twenty held. Six of those are the vendor-reported figures above. The other six rest on a named third party with a published methodology, or on a dated page anyone can open and read, which is what a checkable claim looks like.

Gartner, Magic Quadrant for Personalization Engines, Penny Gillespie, Jason Daigler, Michael Ro, Ross Cosner, 3 February 2026. GARTNER and MAGIC QUADRANT are registered trademarks and service marks of Gartner, Inc. and/or its affiliates. Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. PilotX is not covered in this research and has no commercial relationship with Gartner. All other product and company names are the trademarks of their respective owners.

How do you check a vendor claim in ten minutes?

Work down in this order. Most claims break on the first or second step, and you can do all five before a discovery call without asking the vendor for anything.

What do these checks turn up on PilotX?

Ask every platform on your shortlist for a number measured on your own store, against a control group you set. Ask us the same thing. It is a harder question than asking for a case study, and it is the only one that answers what you actually want to know.

Where to take this next

Checking claims tells you what a vendor has shipped. It does not tell you whether the thing they shipped decides anything per customer, which is the question underneath most of these purchases. The eight platforms that genuinely decide per customer goes through who holds a real decision layer rather than a faster campaign builder, what each one needs underneath it before the first decision gets made, and where the acquisitions went. For the shortlist framed by how autonomously each option actually runs, the agentic platform comparison covers the same field from the buying side.

And if you would rather start from your own numbers than anyone's product page, the Revenue Leak Audit finds the gaps in your retention and puts a money figure on each one. It is free, it takes a few minutes, and it does not require you to believe a single claim on this page.