Every list of AI tools marketers should know starts ageing the day it is published. Products get bought, renamed or quietly retired, and the article stays online recommending something that no longer exists. So rather than write another list, we went back through the tools this piece named originally and checked, in August 2026, which ones are still what they were.
Key takeaways
- An MIT Project NANDA report put 95 percent of enterprise genAI pilots at no measurable return.
- IBM Watson Analytics was discontinued in 2019; the name still circulates in tool lists.
- Albert AI belongs to Zoomd and Dynamic Yield to Mastercard, both since 2022.
- EU AI Act transparency duties for AI systems apply from 2 August 2026.
Start with the number that frames every tool decision
In 2025 a preliminary report from MIT’s Project NANDA, titled The GenAI Divide: State of AI in Business 2025, found that roughly 95 percent of integrated enterprise generative AI pilots showed no measurable effect on profit and loss, with only around 5 percent extracting significant value. The report drew on more than 300 publicly disclosed initiatives, 52 organisational interviews and 153 survey responses collected between January and June 2025.
Two things need saying about it. It is a preliminary, non peer-reviewed report relying partly on self-reported outcomes, so it is one strong signal rather than settled evidence. And its diagnosis was not that the models were bad: the authors point at integration and organisational learning, meaning the failures happened after the tool worked. That distinction is the whole argument for choosing fewer tools and wiring them in properly.
The tools this article named, audited in August 2026
Below is the status of every product named when this piece was first published. Three of the eight are no longer what a reader would assume from the name alone, which is roughly the decay rate we would expect from any tool list left unattended for two years.
| Tool | Status in August 2026 | What a buyer should know |
|---|---|---|
| Jasper | Active, repositioned | Now sold as a marketing agent and content pipeline platform, not a copy generator |
| Copy.ai | Active, repositioned | Moved to go-to-market workflow automation across sales and marketing |
| Google Analytics 4 | Active, now the only option | Universal Analytics stopped processing data on 1 July 2023; the event model is not a rename |
| IBM Watson Analytics | Discontinued in 2019 | Capabilities moved into Cognos Analytics; IBM’s current AI line is watsonx |
| Dynamic Yield | Active, owned by Mastercard | Acquired from McDonald’s, deal closed April 2022 |
| Salesforce Einstein | Active, rebranded around it | Salesforce moved its AI branding to the Agentforce 360 line at Dreamforce 2025 |
| Adext AI | Active | Audience optimisation for paid search and social, unchanged in scope |
| Albert AI | Active, owned by Zoomd | Acquired by Zoomd Technologies in March 2022, now sold as Albert by Zoomd |
Where these tools reliably earn their place
Across the categories above, the applications that hold up under measurement have a common shape: they compress production time on work that was already being done, rather than promising a new source of demand.
- First drafts and variants. Generation tools are strongest at producing the twentieth subject line or the fifth ad variant, where the marginal human cost was high and the marginal creative value was low. They are weakest at the first idea.
- Structuring analytics questions. The value in an analytics platform’s AI layer is usually in surfacing an anomaly worth investigating, not in the recommendation it attaches to it. Treat the second half as a prompt for a human.
- Bidding and audience allocation. Automated bidding genuinely outperforms manual work at volume, because it is an optimisation problem with a clear objective. Below a certain spend it optimises noise.
- On-site personalisation. Worth the integration cost only where you have enough traffic for a variant to reach significance in a usable timeframe. Most sites do not, and running it anyway produces confident conclusions from thin data.
What none of these do is remove the judgement call about whether the campaign should exist. That remains the expensive part, and it is the part the 95 percent figure is quietly measuring.
The compliance layer that was missing from the 2023 tool lists
Two obligations now sit directly on top of this stack, and neither existed in its current form when most AI tool roundups were written.
The EU AI Act entered application in stages. Obligations for providers of general-purpose AI models took effect on 2 August 2025, and the bulk of the high-risk system obligations, along with the transparency duties covering disclosure of AI interaction and the marking of synthetic content, apply from 2 August 2026. For a marketing team the practical consequence is documentation: knowing which model sits behind a tool, and being able to say so.
The second is older and more frequently broken. Any AI-assisted email programme still runs under the consent rules of the market it sends into. Under the GDPR that means a lawful basis and a working withdrawal of consent. Under the American CAN-SPAM Act it means honouring an opt-out within 10 business days and keeping the mechanism live for at least 30 days after the message was sent, with civil penalties running to tens of thousands of dollars per non-compliant email. Personalising a message at scale does not change who was allowed to receive it, a point we go into further in our list of the email marketing mistakes that keep recurring.
What we would tell a team choosing this quarter
Buy for a bottleneck you can name. If nobody in the team can finish the sentence “this tool exists to remove X hours from Y”, the purchase is a bet on a category rather than a fix for a problem, and that is the profile the MIT sample was full of.
Check the ownership before the feature list. Three of the eight products above changed hands or status without changing name, and acquisition usually changes roadmap, pricing and support tier well before it changes the marketing site.
And set the measurement before switching it on, not after. The single most useful thing a marketing team can do with any of this is decide, in advance and in writing, what result would make them turn it off. Almost nobody does, which is the least surprising finding in the whole literature.
Wondering what happens to a brand voice run through all this?
The output problem is not quality any more. It is sameness, and it shows up faster than teams expect.
Sources: MIT Project NANDA, “The GenAI Divide: State of AI in Business 2025”, preliminary report released July 2025, drawing on more than 300 publicly disclosed AI initiatives, 52 organisational interviews and 153 survey responses collected between January and June 2025, for the finding that approximately 95 percent of integrated enterprise generative AI pilots showed no measurable profit and loss effect; the report is preliminary, has not been peer reviewed and relies in part on self-reported outcomes. Google, for standard Universal Analytics properties ceasing to process data on 1 July 2023. IBM, for the discontinuation of Watson Analytics in 2019 and the migration of its capabilities to Cognos Analytics. Mastercard investor communications, for the completion of the Dynamic Yield acquisition in April 2022. Zoomd Technologies, for the acquisition of Albert in March 2022. Salesforce, Dreamforce 2025 announcements, for the Agentforce 360 branding. Regulation (EU) 2024/1689, the AI Act, for the 2 August 2025 general-purpose AI model obligations and the 2 August 2026 application date for high-risk and transparency obligations. United States Federal Trade Commission, CAN-SPAM Act compliance guidance, for the 10 business day opt-out window and the 30 day mechanism requirement. Regulatory dates are stated as published and remain subject to amendment; this article is general information and not legal advice. Updated August 2026.
