This article accompanies the recorded session from the Business AI + Digital Toolbox Programme featuring Giedre Budrius (Forge & Guild) and guest expert Jarred Walker, founder of the marketing agency The Social Subscription.
The conversation covers how a business can use AI tools for social media content without triggering the algorithm penalties now hitting low-effort, obviously automated posts, and gives a practical answer to the question every business owner is asking right now: how much AI is too much?
Which AI and design tools are actually worth using for social media?
The short answer: it depends on volume and complexity, not on which tool is trendiest. Jarred and Budrius compared the platforms their agencies actually use day to day, and where each one earns its place.
Figma versus Canva. Canva remains the default for most small businesses, and for good reason: it’s fast and familiar. But Walker favours Figma for anything that needs to be reviewed as a sequence, such as an Instagram carousel. Canva’s editing workflow is largely page by page, which makes reviewing a full set tedious. Figma works on multiple frames laid out on one canvas, so the entire flow of a carousel or story sequence is visible at a glance. The trade-off is a real learning curve: Walker notes it took his junior staff around a month to get comfortable with Figma, so it’s not a same-day swap for a business used to Canva.
Affinity. Now owned by Canva, Affinity is a cost-effective alternative that covers roughly the same ground as the Adobe Creative Suite (InDesign, Photoshop and Illustrator combined), and integrates directly with brand kits already set up in Canva. Budrius’s advice here is consistent with her broader approach: use what you already have, and only move to a new tool once there’s a clear, specific problem it solves. Switching tools for the sake of switching tools rarely pays for itself.
Claude. Jarred uses Claude for copywriting and to coordinate asset workflows, connecting it to Chrome, Figma, and Google Drive or OneDrive. In one project for client Jim Barry, Claude was connected to OneDrive to search thousands of historical image files by natural language prompt, for example “do you have this photo of Jim Barry from the 1980s?”, instead of manually searching folders. He also shares a prompt pattern developers use to stop Claude or ChatGPT losing reasoning quality over a long chat session: forcing the tool to track results against conclusions with active running notes, rather than letting context quietly drift.
Artlist. Used for AI video generation, Artlist isn’t cheap: plans start around $35 a month and scale up to the $700 a month tier Walker’s agency runs. Pricing is structured around a token-burning model, which caps how much content a single account can generate. That’s a deliberate throttle on the exact problem covered in the next section: unlimited generation is part of what’s flooding platforms with low-value content.
AI-assisted resizing. Figma’s built-in AI agents can automatically resize a finished carousel, in Walker’s example one built around a northern weeds, pests and landscapes campaign, into vertical story dimensions. Budrius flags this as one of the more defensible uses of AI in design: it doesn’t train on or draw from anyone else’s creative work, it simply accelerates the mechanical resizing of content the business already made.
Post dimensions. For Instagram feed posts, Walker recommends tall frames at 1,080 by 1,350 pixels rather than a standard square. A taller frame covers more vertical screen space as a user scrolls, which functions as more advertising space for the same post.
What is AI slop, and why are platforms cracking down on it?
AI slop, what Walker calls the word of 2025, is low-quality, high-volume, automated content clogging up the internet, and platforms are actively working to suppress it. There are three separate reasons driving that crackdown, and understanding all three explains why the penalty isn’t going away.
Storage is a physical cost. Every file, whether it’s an AI-generated video, a photo of a home internet modem, or a “deleted” phone photo that’s actually just compressed and archived, has to be physically stored somewhere. Low-quality automated content is driving an explosion in the volume of data platforms have to hold, and the AI data centres built to store it are a genuine, mounting cost. Platforms are no longer willing to host or distribute content that doesn’t earn its storage space.
It’s linked to attention and wellbeing. Low-effort automated videos, hyper-stimulating clips with rapid cuts and bright colours (talking thumbs, crocodiles, AI-generated fruit, the genre is instantly recognisable), are increasingly described as contributing to “brain rot” and shortened attention spans. Platforms are actively designing their algorithms to purge this style of content on wellbeing grounds, not just cost grounds.
Shadowbanning is real, and it’s enforced. Meta, Google, LinkedIn and YouTube are actively restricting reach for posts that break platform guidelines through lazy, unedited AI content. Walker confirms this directly: shadowbanning isn’t a rumour, it’s an active mechanism. LinkedIn has gone further and deployed a user-facing report option that lets people flag a post as AI slop, which routes it to a review team. And the financial consequences are real too: the popular YouTube channel Capwing estimated to generate around $4.25 million USD a year, had its monetisation suspended, because the storage overhead of that volume of automated content no longer justified the revenue it brought in.
The rule Walker keeps coming back to: it isn’t about avoiding AI, it’s about using it well. A handwritten sketch photographed on paper will outperform a generic, low-effort AI flyer every time.
How do you use AI for content without it getting shadowbanned?
The two case studies from the session show what “using AI well” actually looks like in practice, and both hinge on the same principle: platforms can detect how much genuine editing effort went into a piece of content, not just whether AI touched it at some point.
Case study: the Sunday run club campaign, for clients Tubinarish Cafe and Kapunda Strength Studio. Rather than a single generic prompt, Walker gathered multiple distinct references first: a photo of the cafe’s recognisable storefront tiles, a blurred-motion reference image sourced from Pinterest, and an initial layout built in Artlist. These were compiled into Gemini Omni 3.0 to generate a detailed base image, then refined with follow-up prompts to fix specific errors (an AI-generated figure wearing boxing gloves that had no business being there, for one). Crucially, the raw AI output wasn’t posted as-is: Walker imported it into video editing software, made manual adjustments on his phone (reversing the clip, adjusting speed), and added a custom logo built in Figma. Platforms read the pixel-level edit history of a media file, and because this video carried a genuine multi-stage edit trail, the algorithm recognised human intervention and the post wasn’t penalised. Total time: roughly one to one and a half hours.
Case study: the Jim Barry “Armagh 2023” wine launch. This campaign aimed for a high-end CGI look for a premium wine bottle shot. AI struggled to hold visual consistency across frames, particularly the label text, which kept distorting between frames. Walker spent over fifteen hours prompting and refining before landing on correct panning and macro lighting for the key start and end frames. The AI process cost around $200 in tokens; filming the same macro CGI-style shot in a physical studio with a crew and specialist lighting would have cost thousands. Budrius adds a practical technique of her own here: when generating content featuring her own likeness, she deliberately requests motion blur over her face, which naturally masks AI facial distortion in a way that reads as normal human movement rather than an error.
Both examples point to the same shift in what design skill actually means now. Walker’s summary of it: the core skill is moving away from manual pixel manipulation and toward being highly deliberate, technical and precise with language, because the designer now has to be able to fully articulate what they want rather than simply execute it by hand.
How should a brand structure its content strategy around AI?
Walker’s framework, illustrated using winery client KT Wines, splits content into four pillars, each with a different, deliberate level of AI involvement.
- Human. The people behind the business. AI has no place here: people buy from other people, and this is the pillar where authenticity carries the most weight.
- Product. What the business sells. AI is a legitimate tool here, for styling, layout work, or CGI-style product promotion.
- Location. Where the business operates. Best shot organically, on a phone or with a local photographer, rather than generated.
- Education. What the business teaches its audience. AI is useful here for brainstorming content strategy or outlining a framework, which is then custom-designed manually in Canva or Figma.
Pillars should be reviewed roughly every six months. Walker’s warning: changing a brand’s core messaging too often confuses the audience faster than it refreshes the brand.
Budrius’s framing for the whole session closes the loop with a point she’s made throughout the programme: the biggest mistake a business makes isn’t picking the wrong AI tool, it’s asking the wrong first question. “How should I use AI” is the wrong starting point. “What problems am I actually having, and which of them can be solved with AI” is the right one. That means auditing workflows for what’s slow, repetitive or simple, and automating one thing at a time rather than trying to overhaul everything at once.
On captions and discoverability, the session flags a coming shift from SEO to what Walker calls AIO, AI optimisation. Future text models are expected to embed invisible watermarking in generated text, making purely automated captions easier for platforms to detect. The practical takeaway now: captions need to be genuinely descriptive and keyword-rich so both search engines and social platform AI crawlers can index them properly. A useful diagnostic Walker points to: on mobile, checking a post’s settings (the three-dot menu) will often show the platform’s own AI-generated summary of what the post is about. If the platform can’t generate a summary, the caption likely doesn’t carry enough SEO and AIO signal to be pushed out by the algorithm at all.
