The AI tools we actually use in client delivery (and the ones we tried and dropped)
There is a version of this post where I tell you about fifteen AI tools that transformed our workflow and include affiliate links to all of them. This is not that post. We have tested a lot of tools over the past eighteen months and the honest answer is that most of them saved us about forty-five minutes a week while adding new friction in different places. The net benefit was close to zero.
The ones we kept are genuinely useful. The ones we dropped were either solving problems we did not have or creating dependencies that felt uncomfortable.
The tools we actually use
Claude for writing first drafts of blog posts, ad copy variations, email sequences, and anything where we need to go from zero to something editable fast. Not for final copy, the editing step is still fully ours, but for breaking the blank page. It has probably halved the time our content work takes.
Perplexity for research with citations. When a client asks "what are the benchmarks for email open rates in the Indian education sector," we used to spend an hour piecing together sources. Now that research takes ten minutes, and we can see exactly where the numbers came from.
Midjourney for early-stage mockup imagery when a client cannot provide assets and we need something directional for a proposal deck. We are explicit with clients that these are placeholder images and will be replaced. Nobody has objected.
The ones we dropped and why
Jasper was the first AI writing tool we tried seriously. We dropped it after three months. The output quality was fine, but the interface felt designed for people writing a lot of templated content, and our work is not that. The cost was harder to justify when Claude's output on complex tasks was noticeably better.
An AI SEO content tool I will not name because the team is nice, even though the tool was not good. It generated high-volume blog content very fast. It was also consistently mediocre: grammatically correct, topically relevant, and utterly forgettable. SEO content needs to be good enough for someone to link to it or share it. Competent-but-forgettable does not earn links.
An AI ad creative tool that promised to generate Meta ad variations from a product URL. The images it produced were not usable at a professional level. The copy was serviceable but not better than what Claude produces in a different format.
The pattern in what works
The tools we kept have one thing in common. They are genuinely better than us at a specific narrow task, not sort-of-better at a broad one. Claude is better than us at generating ten variations of an email subject line in thirty seconds. That is a real, measurable advantage. An AI that is "pretty good" at design, copy, strategy, and research is pretty good at none of them in a way that earns its subscription fee.
We are not anti-AI tool. We are anti-mediocre-tool that sounds impressive in a demo.