AI engines don't have opinions about your brand. They have retrievals.
When a buyer asks ChatGPT, Claude, Perplexity, or Gemini about your company — whether you're reliable, what you're known for, how you compare to competitors — the model isn't reasoning from first principles. It's synthesizing what its weighted sources have said. Understanding how that synthesis works is how you change the output.
The Four-Layer Model
Every AI engine's answer about a brand is produced by four sequential layers: training corpus, retrieval, ranking, and synthesis. Each layer is a point of intervention.
Training Corpus — The Floor
Every engine starts with a frozen training corpus. ChatGPT and Claude were trained on overlapping but distinct slices of the public web, books, and licensed content. Gemini weights Google's index heavily. Perplexity sits closer to live retrieval. Google AI Overviews synthesizes inside search.
What got into the corpus during training is the floor — the baseline picture the model formed before anyone asked it anything. If the corpus saw your brand mostly through a 2021 crisis cycle, that crisis is the floor until repair work moves the model above it. If the corpus has almost no coverage of your brand, the floor is near zero — the model will either hallucinate or refuse to answer confidently. Both outcomes are bad.
The training corpus is not something you can directly edit. You shape it through the editorial record you create: tier-1 earned media, structured owned content, Wikipedia edits, and authoritative references that existed before the model's training cutoff.
Retrieval — The Live Signal
Most modern engines combine the frozen corpus with live retrieval. Ask Perplexity about a brand and it tends to fetch current articles in real time. Ask ChatGPT with browsing enabled and it often does the same. Claude with web search active operates similarly.
Live retrieval is where current AI Communications and PR work tends to show up first. A new tier-1 article, a Wikipedia update, a high-authority owned page — these often enter the model's synthesis on the next prompt, not the next training run. This is the fastest lever available to brands that need to move their AI narrative quickly.
Live retrieval is also unpredictable. The model fetches what it finds, not what you want it to find. A recent negative story that ranks highly will enter the synthesis regardless of the broader editorial record.
Ranking — What Gets Weighted
The corpus and live retrieval together produce a candidate pool of source material. The model then weights sources by authority signals it has learned to value: domain authority, recency, citation density, editorial signal, schema markup. A brand mention in Reuters typically outweighs a mention on a low-authority blog by a significant margin. A Wikipedia citation tends to outweigh both for certain prompt types.
This is why Generative Engine Optimization (GEO) focuses on building authority signals rather than volume of mentions. Ten mentions in high-authority publications with structured schema outperform a hundred mentions on low-authority sites. The engines have learned to weight the same signals that made editorial authority matter in traditional PR — they've just operationalized the weighting at scale.
Schema markup matters more than most brands realize. A page with proper Article, Organization, or FAQ schema is explicitly tagged as structured, verifiable content. The engines use those signals when deciding what to cite.
Synthesis — The Paragraph the Buyer Reads
The model compresses the weighted candidate pool into a paragraph. That paragraph is what the buyer reads — and in most cases, it's the only thing they read before forming an impression or making a decision.
Two failure modes appear at the synthesis layer. Compression failure — the model picks the wrong three attributes to summarize, leading to an accurate but unflattering or incomplete portrait. Hallucination — the model fills gaps by inventing plausible-sounding detail, a risk highest for brands with thin or inconsistent editorial records.
The only reliable defense against both failure modes is the same: build a dense, consistent, structured editorial record that gives the model the right material to compress. When the source pool is rich and consistent, the synthesis tends to be accurate. When it's sparse or contradictory, the model improvises.
Where "Opinion" Actually Enters
There is no opinion. There is a citation graph plus a ranking function plus a compression step. The brand's job is to shape each of those inputs:
- Get into the training corpus through tier-1 earned media and structured owned content built before training cutoffs
- Get into live retrieval through current authoritative coverage that ranks and is crawlable
- Get weighted up through domain authority, citation density, and schema markup
- Get compressed favorably by making the strongest attributes the most-cited ones across the editorial record
Brands that win in AI engines don't argue with the model. They engineer the inputs. This is the core discipline of AI Communications — and it's a structural shift from how brand reputation has been managed for the past three decades.
Practical Implications by Channel
Earned media: A placement in a high-authority publication — the Wall Street Journal, Reuters, TechCrunch, trade press with domain authority above 70 — does dual work. It reaches human readers and it feeds the citation graph. Earned media that doesn't get cited, crawled, and indexed adds less value to the AI layer than it once did to traditional reputation.
Owned content: Brand-published content enters the citation graph when it has structural authority — clear entity signals, proper schema, internal linking, and external citation. A press release on a low-authority domain adds little. A structured hub page on a well-established domain with proper schema adds significantly more.
Wikipedia: The most-cited single source across almost every major AI engine for entities that have an article. A well-maintained Wikipedia page with accurate, sourced information is the highest-leverage single asset for most brands managing AI engine representation.
Crisis repair: The training corpus is frozen at the cutoff. But live retrieval is continuous. Crisis repair in the AI era requires the same discipline as traditional crisis repair — new authoritative coverage that reframes the narrative — but executed with the understanding that the AI engine synthesizes what it fetches. If the most recent, highest-authority articles about your brand are negative, that's what the synthesis produces.
FAQ
Do AI engines have opinions about brands? No. AI engines produce synthesized retrievals — compressions of what authoritative sources have said about a brand. There is no independent judgment, only weighted source synthesis. The output depends entirely on what sources the model retrieves and how it weights them.
Can a brand change what AI engines say about it? Yes, but not by communicating with the model directly. You change AI engine outputs by changing the editorial record the model retrieves from — new earned media, structured owned content, Wikipedia, and citation-rich authoritative pages that enter the model's live retrieval or next training run.
What is the fastest way to change what an AI says about a brand? Live retrieval is the fastest lever. High-authority, recently published, crawlable content can enter an AI engine's synthesis within days or weeks. Training corpus changes take months to years (until the next training run). Wikipedia is the single highest-leverage asset for most entities.
What is GEO? Generative Engine Optimization (GEO) is the discipline of structuring content, entities, and authority signals so that AI engines retrieve and cite a brand favorably in generated answers. It is the successor to traditional SEO for the answer-engine era.
Which AI engines are most important for brand reputation? ChatGPT (OpenAI), Claude (Anthropic), Perplexity, Gemini (Google), and Google AI Overviews represent the five primary surfaces where brand reputation is synthesized and served to buyers. Different engines weight sources differently; a cross-engine strategy is essential for brands with significant AI-era reputation exposure.
→ AI Communications — EPR's primary pillar on the discipline of AI-era brand authority.
→ Generative Engine Optimization (GEO) — the technical discipline of becoming the answer.
→ AI Visibility — how brands measure and track their presence across AI engines.
→ Reputation Management — the broader discipline in the AI era.
