Quick answer
Two specific-sounding AI SDR stats circulate right now: a "38-point sender reputation drop within 90 days" and "seven federal and state enforcement actions totaling $24M" against AI-outreach claims. Both trace back to the same single blog post, which cites the reputation-drop figure to an unlinked "Smartlead + Instantly deliverability research" that neither company's own site publishes, and gives the enforcement figure no named cases at all. The real, named FTC actions I could find in the same window (Cox Media Group's $930,000 settlement and Air AI's largely suspended $18M judgment) don't match that breakdown or that total. Treat both numbers as unverified until someone points to where they actually come from.
Two stats worth checking before you repeat them
I'm Hlib Storchak. I build outbound systems for B2B founders and sales teams, and part of that job, with 2000+ meetings booked for B2B clients along the way, is deciding which vendor claims and industry stats are safe to put in front of a client before I've checked them myself. Two AI SDR stats have been showing up together lately, in search summaries and in aggregator roundups, phrased with just enough precision to sound measured: a specific point drop in sender reputation, and a specific dollar total in enforcement fines. Precision is exactly what makes a stat feel safe to repeat. It's also exactly what I've learned to check first.
Claim one: a 38-point sender reputation drop in 90 days
The claim, as it's phrased on Digital Applied's "The Case Against AI SDRs" page, reads: "Smartlead and Instantly 2026 deliverability research shows that domains running AI-SDR outbound at production volume drop sender reputation sharply within 90 days. Median observed drop: 38 points on the major reputation scales." The sentence right after it attributes the cause to email providers detecting "AI-template homogeneity at scale," and the citation line underneath reads "Source: Smartlead + Instantly deliverability research · Q2 2026," with no link to either company.
That's a specific number, a named cause, and two named, credible companies attached to it. It reads like a stat you could put in a slide without a second thought.
Tracing claim one: from a search summary to a dead end
I went looking for the underlying research directly on Smartlead's and Instantly's own blogs first. Neither publishes a "Q2 2026 deliverability research" report with a 38-point reputation-drop figure, or any figure resembling it, anywhere I could find. A general web search for the exact phrase turns up only pages restating Digital Applied's own page, not an independent report at either company. One search result had even attributed the figure to a different site, FirstSales, as its likely origin. I fetched FirstSales' own relevant post directly rather than trust that attribution, and it discusses deliverability decline over a similar three-month window in general terms, but contains no 38-point figure, no reputation-scale number, and no citation to Smartlead or Instantly at all. So the same unsourced number has, at different points, been searchably attributed to two different origins, neither of which actually contains it when you read the page directly. That's not a rounding difference. It's a number with no traceable home.
Why this one is worth flagging specifically. A stat attributed to two real, named companies is more dangerous than an anonymous one, because the names do the convincing for you. Nobody double-checks a number that already comes with "Smartlead and Instantly" attached to it, until they try to find the actual report and discover it isn't there.
Claim two: seven enforcement actions, $24M in settlements
The same Digital Applied page states: "Seven federal + state AG enforcement actions targeting AI-marketing and AI-outreach claims landed in 2025 and the first half of 2026. Settlements totalled $24M," broken down further into "FTC actions ($8.4M), state AG actions ($9.2M aggregate), and class-action suits ($6.4M aggregate)." No case is named. No court, no company, no docket number, no link to a state AG's office or the FTC's own newsroom appears anywhere near the figures.
This is the harder claim to check precisely because it's an aggregate. There's no single source to fetch and compare against, only a total that either matches the real record or doesn't.
Tracing claim two: what the FTC's own newsroom actually shows
I went to the FTC's own press release archive for its 2026 AI-related actions rather than trust a secondhand tally. Two real, dated, named actions fall inside the window Digital Applied's page describes:
- Cox Media Group, MindSift LLC, and 1010 Digital Works LLC, announced May 21, 2026 and finalized August 2026: a combined $930,000 to settle FTC charges that they deceived customers about an "Active Listening" AI-powered ad-targeting service. The service didn't actually process voice data at all; it resold email lists bought from data brokers while marketed as AI-driven conversation-based targeting.
- Air AI and its owners, announced March 24, 2026: an $18M monetary judgment, almost entirely suspended, with $50,000 actually required for consumer relief, over false earnings claims and sham refund guarantees tied to an AI-branded business-opportunity pitch.
That's two real actions, both directly on the FTC's own site, both with names, dates, and dollar figures attached. Neither matches Digital Applied's $8.4M / $9.2M / $6.4M breakdown in size or in category, one of them is a business-opportunity fraud case with an AI label rather than a deception about outreach capability specifically, and neither is a state AG action, which the "$9.2M aggregate" line claims exists. I couldn't find a third, fourth, fifth, sixth, or seventh action anywhere in the FTC's own 2026 AI enforcement listings or in named state AG press releases that would round the count up to seven or the total up to $24M.
Both claims, side by side against what checks out
| Claim | As stated | What checks out directly |
|---|---|---|
| Sender reputation drop | 38-point median drop in 90 days, sourced to "Smartlead + Instantly deliverability research, Q2 2026" | No such report found on either company's own site; the figure only appears restating Digital Applied's page |
| Enforcement action count | Seven federal and state AG actions | Two verifiable, named FTC actions found in the same window; no third confirmed |
| Enforcement total | $24M ($8.4M FTC / $9.2M state AG / $6.4M class action) | Real total across the two verified FTC actions: roughly $18.93M in judgments, almost all suspended; no state AG or class-action component confirmed |
| Underlying mechanism | ESPs detect "AI-template homogeneity at scale" | Directionally plausible and consistent with how spam filters generally work; not itself evidence for the 38-point figure |
What real AI-marketing enforcement looked like in 2026
It's worth being fair to the real cases here, because they're genuinely useful and get buried under the inflated aggregate. The Cox Media Group settlement is the closer analogue to an "AI-outreach claims" case: a company marketed a service as using AI to listen to conversations and target ads accordingly, when it actually just resold broker email lists, and the FTC's own finding was explicit that clicking through a terms-of-service screen isn't opt-in consent for that kind of claim. That's a directly relevant precedent for any AI sales tool marketing a capability it doesn't actually have. Air AI's case is more about business-opportunity fraud wrapped in an AI label than about outreach capability specifically, but it shows the FTC is willing to pursue large judgments, even mostly suspended ones, against AI-branded sales products making earnings promises they can't back up.
Both are worth reading directly if you sell or evaluate AI sales tooling. Neither supports a $24M, seven-action tally.
Why an unsourced number is worse than no number
The mistake I see most often when I take over an account isn't a client repeating an obviously made-up number. It's a client repeating a number that came with two real company names and a specific figure attached, because that combination reads as verified without anyone actually verifying it. A number like "38-point drop" or "$24M in fines" is more persuasive in a board deck or a sales call than a vague warning, and precision gets mistaken for rigor. Once one of these figures ends up in a client's internal risk memo or a vendor evaluation scorecard, unwinding it is a much worse conversation than checking it would have been up front.
The mechanism behind claim one is real, the figure isn't
None of this means AI-volume sending is safe for deliverability, and I don't want this piece read that way. Email service providers pattern-match on structure, not just content, and a fleet of AI-generated sequences that share sentence rhythm, subject-line structure, and personalization-token placement at high volume is a genuinely plausible detection target. I've watched domains lose inbox placement fast when a client scaled sending volume without proportionally scaling warmup and list quality, AI-written or not. What's unverified isn't the mechanism, it's the specific "38 points" figure and the specific two-company citation attached to it. Those are different problems, and conflating them is how an unsupported number survives: the underlying worry sounds right, so nobody checks whether the exact figure attached to it is.
A checklist for this exact flavor of stat
This is the process I actually run before a stat like this goes in front of a client:
1. Find the primary source by name, not by summary. If a stat cites "Company A and Company B's research," go to both companies' own sites and look for the actual report, not a page that mentions the two names.
2. Check whether the citation links anywhere. A named source with no link is a claim to verify, not a citation. Real vendor research usually links to itself.
3. For enforcement or legal claims, go straight to the regulator's own newsroom. The FTC and most state AGs publish dated, named press releases for real actions. An aggregate total with no case names attached is not something you can check against that record.
4. Search the exact figure in quotes and see who else is citing it. If it only appears on pages restating one original source, treat it as that one source's unverified claim, not an independently confirmed fact.
5. Separate the mechanism from the number. A plausible cause doesn't make a specific figure true. Decide what you can defend, usually the mechanism, not the precise number attached to it.
Key takeaways
- The "38-point sender reputation drop" stat traces only to Digital Applied's own page, which cites an unlinked "Smartlead + Instantly" report that neither company publishes.
- The "seven actions, $24M" AI enforcement stat has no named cases attached anywhere; the two real, verifiable FTC actions in the same window total roughly $18.93M in judgments, almost all suspended, and don't match the claimed breakdown.
- Cox Media Group's $930,000 settlement over false "Active Listening" AI ad-targeting claims and Air AI's largely suspended $18M judgment are real, checkable precedents worth reading directly.
- The mechanism behind claim one, ESPs pattern-matching on AI-template homogeneity at volume, is directionally plausible even though the specific figure isn't traceable.
- A stat naming two real companies is more persuasive and less checked than an anonymous one. Check it anyway, especially before it ends up in a client-facing risk memo.
My take
I don't think either stat here started as a deliberate fabrication. This looks like the ordinary way an aggregate claim drifts: someone needed a number for a "the case against AI SDRs" argument, attached two credible company names and a regulator's name to make it feel grounded, and the specifics never got checked again once the post existed. I've unwound versions of this same pattern before, on a reply-rate stat that only half-matched its cited source and on a multichannel lift percentage with no credited source at all. If you need a defensible, actually-verified AI SDR abandonment figure instead of either stat here, Gartner's own June 2025 press release on agentic AI project cancellation is the one I'd point to, because it's dated, named, and directly readable at the source. That's the bar the two stats in this piece don't clear.
FAQ
Where does the "38-point sender reputation drop within 90 days" AI SDR stat actually come from?
It traces to Digital Applied's "The Case Against AI SDRs" page, which cites it to an unlinked "Smartlead + Instantly deliverability research, Q2 2026." Neither Smartlead's nor Instantly's own blog publishes a report containing that figure.
Is the "seven enforcement actions, $24M" AI outreach fines stat real?
No case names, dates, or links accompany it on the page that states it. The two real, named FTC actions I could verify in the same window, Cox Media Group and Air AI, total roughly $18.93M in judgments (almost all suspended) and don't match the claimed breakdown of FTC, state AG, and class-action totals.
What real FTC enforcement against AI-marketing claims happened in 2026?
Cox Media Group, MindSift LLC, and 1010 Digital Works LLC settled for a combined $930,000 in May 2026 over false "Active Listening" AI ad-targeting claims. Air AI settled in March 2026 with an $18M judgment, almost entirely suspended to $50,000, over AI-branded business-opportunity fraud.
Does high-volume AI SDR sending really hurt sender reputation?
The mechanism is plausible: email providers pattern-match on structural similarity across sequences, and high-volume AI-generated sending at scale is a reasonable detection target. That doesn't make the specific 38-point figure verified. Treat the risk as real and the exact number as unconfirmed.
How do I check a stat like this before I repeat it to a client or a boss?
Go to the named sources' own sites directly rather than a summary, check whether the citation actually links anywhere, check regulator enforcement claims against the regulator's own newsroom, search the exact figure in quotes to see if it only restates one origin, and separate a plausible mechanism from an unverified specific number.