Generative AI shifted in several important ways this quarter, and each one affects how organisations are found, understood and evaluated inside LLMs. New evidence shows what models actually read, the model race is reshaping which systems interpret your brand, and platforms like OpenAI are moving closer to AI-driven commerce. At the same time, device ecosystems and enterprise tools are building AI into everyday workflows. For PR and comms teams, these changes aren’t abstract; they reshape discovery, influence and measurement. Here’s what stood out in Q4 and how it reshapes the work ahead.
What LLMs Actually Read
One useful Q4 update comes from Muck Rack’s latest Generative Pulse report, What Is AI Reading? The December 2025 analysis of more than 1M links shows that journalism and earned media still dominate AI citations (around a quarter of links are journalistic and over 80% are earned), AI strongly prefers fresh coverage published in the last year, and press releases are now cited 5× more often than they were in July. The study also finds that outlet authority and niche, industry-specific sources heavily shape which brands surface in generative answers, and that models are constantly adjusting their citation mix.
ChatGPT Ads: Hype Now, Reality Later
Over the last few weeks, rumours of “ChatGPT ads” have exploded, fuelled by leaked Android app code referencing an “ads feature” and search-style sponsored cards. At the same time, OpenAI leaders are clear: no live ad tests are running today, and any future ad product will be carefully designed to protect trust and usefulness. With Sam Altman delaying formal ad initiatives to focus on the core ChatGPT experience, the takeaway for comms teams is simple: ads inside generative chat are coming into view, but they are not here yet. Now is the moment to scenario-plan how a paid visibility layer in ChatGPT could shape discovery, GEO and media planning, long before the product is live.
Gemini 3 Ups the Stakes in a Multi-Model World
The most high-profile twist in the Q4 model race is Google’s Gemini 3 pulling ahead of ChatGPT on key intelligence benchmarks. According to The Guardian, Sam Altman has declared a “code red” internally after Gemini 3 “outperformed rivals on various benchmarks,” warning staff that the launch could create “temporary economic headwinds,” even as ChatGPT serves 800 million weekly users. For PR and comms teams, the message isn’t to switch tools overnight, it’s to recognise that we’re now operating in a genuinely multi-model environment. Gemini looks strong on reasoning and multimodal tests; ChatGPT still leads on adoption and usability. The next practical step: pilot Gemini alongside ChatGPT in research, content and measurement workflows, and assume that “what AI sees” about your brand increasingly depends on multiple ecosystems.
Apple’s ‘Quiet AI’ vs Samsung’s ‘Everywhere AI’
Apple has been criticised for moving slowly on AI compared to rivals, but 2025 saw Apple Intelligence expand across iPhone, iPad, Mac and other devices, with capabilities like on-device writing assistance, translation and image generation. Apple’s focus remains on privacy and clean system integration rather than flashy experimentation, and new leadership in the AI group signals a push to accelerate. For comms teams, this means Apple users are steadily gaining subtle but powerful AI assistance inside tools they already rely on, with minimal disruption to established workflows.
Samsung has taken the opposite path, positioning Galaxy AI as a flagship differentiator and pushing it onto hundreds of millions of devices, from phones to TVs. Its generative features are more visible and experimental: smart editing tools, conversational assistants and context-aware suggestions embedded across the interface. For PR and comms teams, this signals a user base increasingly comfortable using AI in real-time translating, summarising and redesigning content on the fly. Expectations for speed, localisation and creative output will rise accordingly.
Claude 4.5: The Agentic Co-Worker Arrives
Anthropic has spent Q4 consolidating its Claude 4.5 family and pushing hard into enterprise use cases. After Sonnet 4.5 and Haiku 4.5, November brought the debut of Claude Opus 4.5, its most capable model yet for coding, multi-step agents and computer use. Opus supports longer-running “agents” that can work across documents, slides and spreadsheets, and now integrates directly into Chrome and Excel. For PR and comms teams, Claude is no longer a “chatbot alternative”, it’s becoming an AI co-worker embedded inside everyday tools, raising expectations for accuracy, automation and speed across content, research and reporting workflows.
When Earned Media Turns Into Commerce
OpenAI’s move into e-commerce is one of Q4’s most important shifts for PR and comms. With ChatGPT now offering native checkout and connecting to more than one million merchants through the Agentic Commerce Protocol, earned media becomes directly actionable. Reviews, roundups and product mentions already fuel most LLM citations; now they can also drive instant purchases. Analysts describe this as the emergence of an “agentic commerce layer,” where AI assistants guide discovery and complete transactions end-to-end. For comms teams, this unlocks clearer attribution (media-driven queries that convert), new pitching strategies (which mention types convert best) and expanded ROI metrics, including checkout volume and per-mention revenue impact. Earned media is no longer just credibility; it’s becoming a driver of demand.
Reducing Hallucinations in Daily Comms Workflows
AI hallucinations remain one of the biggest operational risks for PR and comms teams. Even advanced models can invent quotes, misstate timelines or fabricate data: errors that escalate quickly. The most reliable safeguard is to design prompts and workflows that limit the model’s freedom to guess. That includes grounding every task in approved materials (press releases, bios, product sheets), explicitly restricting it from using general knowledge, and giving clear permission to say “I don’t know” when information isn’t provided. For sensitive content, shift from open-ended “create” prompts to “summarise this” or “reword this approved text,” and ask the model to flag uncertainty or cite sources. Finally, check factual elements separately from tone: hallucinations hide in the details, not the phrasing.
What to Watch in Q1 2026
Early 2026 is already signalling another round of changes. OpenAI is expected to widen its commerce integrations, potentially turning ChatGPT into a default shopping layer for more categories. Google’s next Gemini updates aim to strengthen multimodal reasoning and structured retrieval, while Apple is preparing the next step in Apple Intelligence, likely expanding support to more devices. Regulation will tighten in the EU and UK around data provenance and transparency, with implications for how earned media is surfaced inside AI systems. And across all major platforms, assistants are set to take on more agentic tasks, moving from answering questions to completing workflows end-to-end.
For comms teams, the signals coming out of Q4 and into Q1 point to a discipline where precision, clarity and evidence carry more weight than volume. This is the moment to strengthen the inputs that AI relies on. Because the systems reading your brand are getting faster, not simpler.
Leave a Comment