7 min read AI D2C

The AI Consolidation Nobody's Talking About: What October's Data Tells Publishing and EdTech

The latest Similarweb Global AI Tracker data reveals something most publishers and educators will misinterpret:

Standalone AI content tools are declining sharply

  • Writing tools down 15%

  • Image generation down 9%

  • Voice generation down 12%

The temptation is to breathe a sigh of relief: perhaps the AI disruption fears were overblown after all.

That would be precisely the wrong conclusion.

The Real Story in the Numbers

Whilst specialist content generation tools decline, general AI platforms are consolidating extraordinary dominance.

ChatGPT maintains roughly 65% market share whilst growing 5%. Claude shows 47% growth. Gemini, backed by Google’s resources, is up 64%. Perplexity, integrating AI into search, has grown 66%.

This isn’t users abandoning AI. This looks like market professionalisation.

The amateur content flood is plateauing, but professional deployment using sophisticated, integrated workflows is just beginning. The standalone tool market is dying because capability is being absorbed into platforms that handle multiple content types seamlessly.

Anyone who’s watched platform economics knows what this means: when technology democratises and consolidates, competitive advantage moves elsewhere. The question facing publishing and education isn’t whether AI disruption is coming - it’s whether we understand where value actually lives once baseline AI capability becomes ubiquitous.

What Publishers Need to Notice

Three signals in this data:

First, the voice generation plateau. Traffic to voice AI tools like ElevenLabs may be down 17%, but this reflects integration into workflows rather than abandonment. Voice AI quality has reached “good enough” for many applications. The technology isn’t retreating, it’s becoming infrastructure. Audiobook production economics could face structural pressure within 18 months, regardless of whether we’re ready for that conversation.

Second, the freelance platform decline. Digital freelance services down 7% year-over-year doesn’t sound dramatic until you notice it’s steady and sustained. This creates dual pressure: downward cost expectations alongside (potentially) diminishing quality benchmarks. The risk isn’t that all freelance work disappears - it’s that “good enough at lower cost” becomes the new normal before we’ve articulated what premium quality actually means in an AI context.

Third, the search displacement signal. Perplexity’s 66% growth represents something more fundamental than a new search engine. If readers increasingly ask AI “what should I read after X” rather than browsing Amazon or bookshops (or our own D2C stores), the entire discovery mechanism shifts. Metadata strategies optimised for traditional search and recommendation engines may be solving yesterday’s problem.

The pattern across all three: the change is already happening at infrastructure level, not at the edges.

EdTech’s More Acute Timeline

The education sector data is starker. Consumer EdTech shows steady 9% year-over-year decline, but this masks catastrophic collapses in specific categories:

  • Chegg (homework help): -66%

  • Coursehero: -60%

  • Mathway: -61%

These aren’t corrections. These are structural shifts.

More concerning for language learning specifically: Duolingo is down 17% with a falling trend (getting worse, not stabilising), despite being AI-enabled for years. Babbel is down 39%, though showing signs of recovery. These are established players with sophisticated technology, strong brands, and millions of users.

If AI-enablement alone were sufficient protection, they wouldn’t be declining.

The homework help collapse tells us AI is already extensively deployed by students, regardless of institutional policy. The language learning pressure suggests that when general AI platforms offer conversational practice for free, even sophisticated adaptive learning systems face margin compression.

“All other” in EdTech is up 9%, indicating new entrants or different approaches are gaining whilst established players struggle. The market isn’t shrinking - it’s reallocating to solutions that work differently.

The Uncomfortable Question About Value

Here’s what makes me uncomfortable about this data: it suggests that technology moats are eroding fastest in precisely the areas where education and publishing have been investing heavily.

Adaptive learning? ChatGPT offers that conversationally. Personalised practice? Claude will generate exercises endlessly. Content at scale? Trivially easy now. Multimedia production? Increasingly commoditised.

If the baseline capabilities that justified premium pricing become freely available, where does value actually live?

Three possibilities, none of them comfortable:

Structured curriculum and expert pedagogy - but only if demonstrably superior to what users can cobble together themselves. “Trust us, our approach is better” isn’t sufficient if learners can’t perceive the difference.

Credentialing and certification - but only if employers and institutions continue to value the credentials. Market value of credentials is itself under pressure as AI makes skills testing more accessible.

Community, motivation, and accountability - genuinely hard to replicate, but also hard to monetise at scale. The most defensible value proposition may be the least scalable.

Or perhaps value is shifting somewhere we’re not looking yet. Maybe the true situation is: we don’t actually know.

What the Data Doesn’t Tell Us

This is traffic data, not usage depth. Claude’s 47% growth is significant, but Similarweb doesn’t capture API usage or enterprise deployment. Professional adoption via API almost certainly exceeds these consumer traffic figures, possibly substantially.

We’re reading historical trendlines and extrapolating, which is inherently uncertain. Markets often move slower than data suggests, but occasionally much faster.

Publishing and EdTech appear in the “disrupted sectors” analysis based on historical patterns, not predictive modelling. We’re not automatically disrupted, we’re vulnerable to disruption if we don’t respond thoughtfully. Crucially different.

The Design Problem Nobody’s Solved

Design platforms like Canva are up 22% whilst AI image generation tools decline 9%. This tells us something important: designers are using AI, but consumers aren’t replacing designers. The tool augments; it doesn’t replace. Yet.

This suggests a possible pattern: AI handles commodity production whilst humans focus on creative direction, strategic thinking, and integration.

If this pattern holds, the challenge isn’t AI replacing humans, it’s that the work requires different humans doing different things - somthing that has been said for some time, but here is data backing this up. That’s an organisational design problem, not a technology problem. Organisational design problems are harder to solve because they involve people, power, and identity, not just capabilities and tools.

What Would Strategic Response Actually Look Like?

I suspect success won’t come from either resisting AI or adopting it uncritically. Success will come from identifying what humans do that AI can’t (or shouldn’t), and organising the entire operation around that distinction.

That’s philosophically simple but operationally brutal. It requires honest assessment of current capabilities, willingness to abandon approaches that worked historically, and tolerance for uncertainty about what will work in future.

It also requires speed. The data suggests 12-18 months before multiple areas face structural pressure simultaneously. Voice, design, content creation, discovery - all shifting now, not in five years.

But I could be wrong about timing. Markets often move slower than data implies and Gartner’s AI hype cycle (this link to a different article if you want to see a non-pay walled version) might provide some comfort (and warning). The risk is that “could be wrong about timing” becomes excuse for inaction, and we discover timing was right but response was too slow.

The Bit That Keeps Me Up at Night

The decline in content generation tools isn’t evidence that AI fears were overblown. It’s evidence the market is professionalising. Amateur hour is ending. The serious deployment phase is beginning.

The publishers and educators who navigate this successfully won’t be those who resisted AI or those who adopted it uncritically. They’ll be those who figured out their actual value proposition in a world where baseline AI capability is free and ubiquitous, and reorganised their entire operation around delivering that value.

Most organisations will move too slowly. Not because they’re incompetent or unaware, but because organisational inertia is real, difficult conversations are postponed, and the pressure to maintain current operations overwhelms the imperative to prepare for different ones.

The question isn’t whether disruption is coming. The data shows it’s here. The question is whether we’ll respond whilst we still have agency in how we adapt, or whether we’ll be forced to respond reactively when we have fewer options and less time.

That uncertainty is uncomfortable but honest. What I do know is that pretending we have more time than we do is the surest way to run out of it.

Views expressed are my own and based on analysis of publicly available market data, analysed with the help of AI. They do not represent the strategy or policy of any organisation I’m associated with.