TL;DR: AI implementation takes time and experimentation. We’ve tested custom GPTs, AI-generated images, and SERP analysis tools to find what works for SMEs. The result? Faster content creation, better brand consistency, and lower costs, but only after honest trial and error.
What you need to know:
- AI tools need refinement over time, not instant setup
- Custom GPTs save days on content while improving consistency
- Human judgment remains essential for quality control
- AI helps uncover human questions behind keyword data
- Experimentation costs have dropped, making testing easier
Why AI Implementation Takes Longer Than You Think
Here’s what most agencies won’t admit: we spend a lot of time messing around with AI.
Not carelessly. Necessarily. Just as we experimented with Screaming Frog and SEMrush in their early years, we’ve had to experiment with AI.
We assumed AI would save time immediately: set it up, let it run, move on. That’s how most software works once you master the interface.
AI is different. Large language models require ongoing refinement and prompt tuning before they reliably deliver. In McKinsey’s recent survey, 78% of organisations reported using AI in at least one business function, up from 72% earlier in 2024. McKinsey & Company
Similarly, Salesforce reports that 75% of SMBs are at least experimenting with AI, with 83% of growing businesses increasing their investments next year. Salesforce
The process we use now after months of testing produces consistent quality faster. But getting there required investment most agencies won’t admit to their clients.
Bottom line: AI setup is an investment, not a quick win. Honest experimentation beats false confidence.
What Do SMEs Actually Want From AI?
SMEs don’t want an AI consultant selling them solutions. They want a partner who will figure this out with them.
The question we hear most: “What should we be doing?”
Not “Can you implement AI?” but “What’s worth implementing?” Because AI is everywhere now, and the noise makes it impossible to know what matters.
That’s where we come in. We test tools like Pressmaster (for interview based thought-leadership content generation), bespoke software that analyses the top 20 SERP competitors and maps content pathways, and build custom GPTs trained on client brand voices. Then we tell clients which ones actually move the needle.
Yesterday, we put six AI-generated content topics in front of a client. They rejected one. The other five were spot on. That’s an 83% hit rate, achieved in a fraction of the time traditional discovery would take.
That changes how clients view experimentation. SMEs don’t need 50 AI tools, they need curated solutions. Partnership beats consultation.
How Custom GPTs Save Time While Improving Quality
We had a client doing a global content migration project. The acquired entity had 20 case studies, rich content but inconsistent in tone and brand alignment. Rewriting them the traditional way would have taken a week with multiple writers and inevitable voice drift.
Instead, we built a custom GPT in 15 minutes, training it on the client’s existing approved case studies. We fed it the 20 new ones.
Result: A day and a half of work instead of a week. And more importantly: the client said the rewrites were better than the originals.
Not just consistent, improved. The GPT had internalised what “right” looked like for that specific brand and applied it uniformly.
If the tone wasn’t quite right, we adjusted it across all 20 pieces in one go. That flexibility matters for SMEs trying to move fast without sacrificing quality.
What this means: custom GPTs trained on approved content deliver consistency at scale. Fifteen minutes of setup can save days of work.
Where Does Human Judgment Matter Most?
We ran an AI-generated campaign for Springfield Healthcare. The original idea: children dressed as elderly people, sneaking into their grandparents’ care home, a highly conceptual, emotionally sensitive concept. The estimated shoot cost was £67,000.
We generated all images via AI instead. But many subtle decisions still required human judgment.
A creative director looked at those images and knew immediately when something crossed from “good enough” to “this will hurt the brand.” That judgment doesn’t come from a prompt, it comes from years of experience.
Prompt engineering might get you 80% of the way there. The remaining 20%, the nuance, authenticity, brand safets, depends on human refinement.
Photoshop skills, design instincts, brand sensitivity: all more important than ever.
The question we’re wrestling with now: how do you teach that eye to apprentices raised in AI-first workflows, rather than the other way around?
The reality: AI generates raw material. Humans decide what’s publishable. That expertise remains irreplaceable.
How AI Reveals the Human Side of SEO
Traditional keyword research focused on volume, difficulty, and SERP features. We would spend days analysing data and then build strategy on what ranked not what people actually wanted to know.
Now we run that same analysis through an LLM, asking: “What questions are people asking? What problems are they trying to solve?” The AI mines Reddit, Quora, People Also Ask, and forum threads surfacing real human intent. We then present those questions to client and ask: “Which of these should we prioritise?”
The machine helps us uncover human questions behind keyword data.
That’s the irony no one admits: AI isn’t making SEO less human it’s making it more human. By surfacing the questions people really care about, the content becomes more useful, not just optimised.
We still carry out technical analysis. But now it’s filtered through a layer of human intent which means we start closer to what the reader actually wants.
What Good SEO Means in 2025
SEO has been trending toward human-first content for years. But the tools to deliver that at scale didn’t exist…until now.
We ask:
- Will a real person want to read this?
- Will they find it useful?
- Are we genuinely helping someone in their journey?
Those questions always mattered, but often got buried under density metrics, keyword stuffing, and SERP feature chasing.
Now AI handles the technical legwork. We focus on strategy and value.
Across clients, we often generate a quarter’s worth of content topics in one afternoon. Clients reject maybe 15–20%. The remaining ideas become the roadmap.
Effect: consistency improves, speed increases by 400%, cost per piece drops roughly 50% — but only for those willing to go through the messy early phase.
In 2025, good SEO means answering real questions faster. AI handles technical analysis so you focus on helping people.
Why the Cost of Experimentation Has Dropped
The cost of experimentation has dropped dramatically, not just in money, but in risk, time, and failure tolerance.
If a batch of six content ideas yields one miss, that’s acceptable. We haven’t wasted a week. We spent a few hours, got feedback, and iterated.
That changes client dynamics. They’re more willing to test offbeat angles or “what if” ideas they otherwise would’ve feared.
Because when something fails, you’ve lost very little. You adjust, you learn, you try again.
That’s what “AI-forward” means: not replacement, but expansion of capacity to test, learn, and grow, faster and more cheaply than ever.
Even so, the experienced eye remains the filter. Strategy remains human. The client relationship remains about trust and honesty. But we move faster, test more, and converge more quickly on what resonates.
Most agencies hide their mistakes. We don’t, because that’s how you build trust and resilience in 2025.
The Simple AI Tasks Saving So Much Time
| Task Type | Traditional Workflow | AI-Enhanced Workflow | Time Saved | Quality Impact |
|---|---|---|---|---|
| Content Ideation | Keyword research, competitor review, manual topic mapping (approx. 2 days per quarter) | AI SERP analysis + client validation (0.5 day) | 75% faster | More relevant to audience intent |
| Case Study Rewrite | 20 case studies rewritten manually by 2 writers over 5–7 days | Custom GPT trained on brand voice – rewritten in 1.5 days | ~4.5 days saved (65%) | Improved tone and consistency |
| SEO Research & Planning | Manual keyword clustering, spreadsheet mapping (2–3 days) | LLM-driven analysis of 20 SERPs + auto clustering (1 day) | 60–70% faster | More human-focused topic insights |
| Ad Image Production | Studio shoot, 8 actors, lighting, post-production (~£67,000 budget, 3 weeks lead time) | AI-generated imagery + Photoshop refinement (2 days) | >90% faster | Comparable creative quality at lower cost |
| Quarterly Content Strategy | Manual brainstorm, internal reviews, presentation (4–5 days) | AI-assisted topic generation, human refinement, client feedback loop (1.5 days) | ~70% faster | Higher output and alignment |
Common Questions About AI in Marketing
How long does it take to set up AI tools properly?
Months, not days. Expect continuous refinement and tuning.
Will AI replace human marketers?
No. AI gives you drafts and data — humans provide strategy, quality, and judgment.
What’s the most common agency mistake?
Claiming they’ve “nailed it.” The better move is transparent experimentation.
Should SMEs wait until tools mature?
No. 88% of marketers already use AI in their roles, per SurveyMonkey’s 2025 stats. SurveyMonkey The risk is falling behind.
How do you maintain brand voice with AI content?
Train your GPT on approved content. Have it learn what “good” looks like — then apply it consistently.
How do you know AI-generated content is good enough?
Experience and quality filters. You don’t publish blind.
The Takeaway
- AI doesn’t replace creativity it amplifies it.
The difference between hype and impact lies in honest experimentation. We combine tools, tests, and human curation so SMEs don’t get lost in the AI noise. Instead, they get progress steadily, transparently, and humanely.
