Content marketing can be magical with both an AI and a Human in the Loop
ChatGPT, Claude, Gemini and other large language models by their creator’s own admissions have a long way to go in reproducing human conversation. But let’s not kid ourselves: these are powerful tools and there is no putting the genie back in the bottle.
Case in point: this is how an early version of ChatGPT suggested I begin this blog: “As businesses continue to navigate the ever-changing landscape of digital marketing (already falling asleep) , the importance of public relations and content marketing agencies cannot be overstated” (that was a painful few seconds, I’m sorry to put you through that).
The world’s most advanced artificial intelligence keeps proposing, despite my repeated objections, that I must end most pieces of writing with “in conclusion…” — ouache! It’s an incredibly advanced research and translation tool, but in terms of replacing human leadership or even natural conversation, it’s not quite ready for primetime.
And since LLMs are a summary of popular knowledge, managing innovative ideas and unconventional leaders are not core capabilities.
That doesn’t mean we can afford as creatives to ignore AI or take a militant stance against it, which would be foolish. As per ChatGPT, it can “automate certain tasks, generate ideas, and even improve writing style” — that’s true, it can help by doing those things.
Should you incorporate AI tools into your communications processes, whether you are an individual, SMB, agency or marketing department? Absolutely, with caution. With skilled prompt-writing and editorial rigour, AI tools can reduce production time and even increase the quality of communications.
What AI can’t do is your job for you. At least not yet.
Here’s how it can help you, and how TNKR Media uses AI throughout our processes.
Translation
Our position on automated translation has changed substantially.
High-quality AI translation paired with a skilled bilingual editor can now produce excellent business communications considerably faster than traditional workflows alone.
We still assume human review will be required, particularly where tone, technical terminology, Québec usage or political and cultural context matter.
Revision
We’re a long way from Clippy, Microsoft Word’s anthropomorphic paperclip. Automated grammar and style tools have become standard parts of professional editing.
They are particularly effective at detecting repetition, inconsistencies, awkward syntax and mechanical errors.
But editorial judgment remains important, especially in a Public Relations (PR) or Public Affairs (PA) context that may be too new, innovative or granular for AIs to manage accurately.
Transcription
Transcription has been one of the most consequential AI applications in our production process.
A recorded conversation can now become searchable text within minutes, allowing producers to identify arguments, quotations and themes without manually transcribing an entire interview.
For our Content Capture model, that is a major efficiency gain: the executive speaks naturally while the production system organizes the raw material immediately afterward.
Human review remains necessary, especially with names, technical terminology and multilingual conversations.
Research
Large language models increasingly function as a layer above traditional search. They can rapidly identify themes, generate research pathways, compare documents and organize large quantities of information.
They can also be wrong. For professional communications, AI research should therefore accelerate verification, not replace it. Important claims still need reliable sources and human judgement.
Logistics
Scheduling, filing, converting, transferring, publishing and connecting platforms are exactly the kinds of tasks we want technology to handle.
The objective is to automate administrative friction so that human time can be concentrated on interviews, analysis, strategy and editing.
Design
Generative systems have also accelerated routine visual work. They can produce concepts, variations and supporting graphics quickly, while human producers determine which visual direction actually serves the subject, audience and brand.
The same distinction applies here as in writing: execution can increasingly be automated; Innovativeness, taste and accountability are harder to delegate.
Drafting: AI Accelerate vs. Original Content
Since our last AI policy update in May 2023, we have decided to approve AI-generated text for clients who request it, under specific circumstances.
Previously, as recently as one year ago, there were no significant time savings between AI-generated text and our Original Content process, which has for many years used AI in aforementioned secondary roles: transcription, translation, logistics, etc. Today, LLMs have become better at generating text but we still maintain strict controls on leadership-oriented writing.
AI Accelerate is now available to TNKR Clients requesting automated blogging in Consumer Blog formats, but it not available for Feature Blog formats which continue to use AI tools in secondary roles.
Unless you cannot compete in your field without them, try to avoid blog automation in favour of smaller batches of content that are rich in human insight.
The Principle: Human in the Loop
Automating generic information is efficiency. Automating someone’s supposed expertise is something else. TNKR allows AI-assisted drafting for clients who want it under our Accelerate mode, with guardrails.
Since we founded TNKR Media, we’ve considered emerging AI ethics guidelines like the Montreal Declaration and traditional media codes of ethics like that of the Canadian Broadcast Standards Council (CBSC) to help inspire our own efforts to produce media responsibly.
The Human in the Loop (HITL) principle is not only useful but baking it into your processes is a safe way to make sure AI doesn’t mess anything up. In its simplest form, the concept simply implies that we ought to help “the computer in making the correct decisions in building a model,” and not let it take the lead.
Finding the right ‘prompts’, combinations of requests to solicit responses from the AI, is the most difficult part of using these tools. On that, I’ve found helpful advice from Mr. Prompts, Rowan Cheung and Max Rascher, among others.
When we attempted to have GPT write a column in my voice, the results weren’t impressive. After about an hour of experimenting with different prompts to obtain a fairly mediocre result, I might as well have buckled down and written a rough draft, or freestyled one as a voice note.
The tools continue to improve considerably, but the HITL principle can’t change for the foreseeable future. AI can accelerate research, organization, drafting and production; it cannot supply the experience, judgment or original point of view that makes expertise worth publishing in the first place. The best way to future-proof communications work is to use automation where it creates efficiency while protecting the human contribution where it creates value.
That means keeping humans in the loop, and an expert at the centre.
And that summarizing point, to be completely transparent, was suggested by ChatGPT. It is aware of its limitations, as we should be.
