Written by Sophie Babic

Artificial Intelligence (AI) is moving fast, shaping industries and sparking debate. For some, it’s a powerful tool for creativity and efficiency. For others, it brings uncertainty.  

October marks one year since Google began using AI Overviews in Australia, which has seen a tangible impact on traffic. 

At IMPACT, we continue to unpack how AI influences visibility, credibility, and performance across channels, and what brands can do to stay ahead. 

In this blog, we’ll take you through the most commonly asked questions and misconceptions we hear about AI. 

Are LLMs reliable? 

AI chatbots have been known to generate false or misleading information at some point. If you’ve had one of IMPACT’s AI Search and Performance Benchmarks done, you’ll know some platforms hallucinate, producing confident, believable text that’s factually incorrect.  

Many large language models (LLMs) have a knowledge cutoff, meaning they were trained on data up to a certain date and can’t access real-time information unless connected to a live search tool. 

Human oversight remains essential to ensure factual accuracy, context, and cohesion. 

What sources do LLMs prioritise most? 

LLMs favour content from reputable publishers, established news outlets, and academic or government sources due to their perceived authority. Studies show AI search systems give strong weight to high-ranking Google pages when generating answers. For example, a recent study of searches found models, like ChatGPT and Perplexity, often pull from top-ranking sources. 

Why am I served media stories as sources to my answers? 

When generating an answer, the LLM prioritises sources based on their relevance and credibility to the query, which includes licensed partners like the media, but is not limited to them.   

Media partnerships primarily serve to provide LLMs with legal access to high-quality, authoritative, and paywalled content, helping to improve accuracy and trustworthiness.  

Media partnerships in Australia include ChatGPT with NewsCorp, and Gemini/Google overviews with AAP). 

The platforms are initially trained to scan a massive, diverse foundation of data, including media plus the open web, reviews, books, Wikipedia, and forums. 

Does AI rank advertorial content differently than editorial content? 

AI systems and LLMs tend to favour editorial content over explicitly sponsored material.  

Editorial-style content aligns more closely with factual reporting, the type of information LLMs are trained on, while advertorials are seen as more promotional and less objective. 

While certain prompts (e.g. questions) sometimes surface “crawlable” sponsored content that directly answers the prompt, editorial integrity still wins when it comes to perceived credibility and visibility.  

Do LLMs pull from influencer content and videos? 

To a degree, yes. LLMs primarily work with text, so they can’t “watch” a video visually (yet!). Instead, they ingest text derived from video content, such as captions, transcripts, titles, descriptions, and comments.  

Influencer blogs and social media posts are also easily ingested because they are purely text-based. For video content to influence an LLM’s knowledge base or surface in an AI search result, creators must ensure their videos include high-quality, relevant text data, as this is the digital material the AI actually learns from. Influencer content also acts as an additional, third-party touchpoint for LLMs to inform answers. 

Do ChatGPT and similar tools crawl my social media?  

Not exactly. Most LLMs and major AI models do not actively scrape private or login-protected social content. They primarily rely on public or licensed data for training and inference. 

However, in certain contexts, the boundaries are blurrier. Meta (owner of Facebook and Instagram) is already using public platform data to train AI models, in some cases going back as far as 2007.  

Posts, photos, comments, captions, and other content you’ve shared publicly can be used under Meta’s terms of service and used in AI model training. Private or restricted content is off limits. Anything behind login walls or privacy settings (e.g. private Instagram accounts, Direct Messages, private LinkedIn posts) should not be accessible to these models. 

Are AI search results the same for everyone? 

Not quite. Results can vary based on location, language, user behaviour, search history and query phrasing and context. 

A 2024 arXiv study found that higher degrees of AI personalisation can reduce factual consistency so the more the model “adapts” to you, the more its output may drift from objective truth.  

Privacy and data regulations in different countries also impact what information LLMs can access, further diversifying results.  

Do I need to state when AI has been used? 

Yes, it should be clearly stated if you’ve used AI to create content for trust, transparency, and compliance reasons. 

A recent Sprout Social survey found that audience trust is significantly higher when AI-generated content is disclosed. Ethical guidelines and emerging platform policies are increasingly urging or requiring disclosure practices. 

If you’re interested in leveraging AI as part of your communications strategy, contact Nicole Webb at nicole@impactagency.com.au.