Can You Control What AI Models Say About Your Company?

By EPR Editorial Team

Originally published June 2026. Updated June 2026.

You can't directly edit what ChatGPT, Claude, Perplexity, Gemini, or Google AI Overviews say about your company. You can edit the citation graph they retrieve from. Wikipedia entries, tier-1 earned media, original research, and structured owned content are the four inputs that move AI engine outputs. Everything else — paid placements, SEO blog farms, low-authority press — has generally negligible effect.

The short answer: not absolutely, but more than most CMOs assume.

The longer answer is the difference between brands building citation share and brands inheriting their own reputation. Across ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews, what an engine says about a company is downstream of a citation graph — the assembled set of training inputs, retrieved sources, and structured signals the engine reaches for when a buyer asks. The citation graph is editable. The engine's output is not directly editable. That distinction is the entire operating frame.

Part of EPR's AI Communications coverage. See also: Wikipedia Is Now Investor-Grade Infrastructure · The New Rules of AI-Readable Disclosures · When AI Defames You.

What you can't control

The engine's underlying training data. Once it's trained, it's trained. New versions update, but you can't reach into the corpus.

What competitors publish about you. Free press, free criticism — both enter the citation graph.

What users say in reviews, forums, and social posts. These are increasingly retrieved, especially by Perplexity.

The engine's ranking function. The proprietary weights that decide which sources rank highest are not editable.

The vendors selling "guaranteed AI placement" are selling something the engines don't take. The engines don't accept requests.

What you tend to be able to influence

The volume and authority of sources that mention you. Tier-1 earned media, Wikipedia presence, original research. Often the single biggest lever.

The schema and structure of your owned content. Engines parse structured pages cleanly. Unstructured ones tend to get ignored or hallucinated around.

The recency of the citation graph. A brand whose most-cited sources are five years old often has a stale AI reputation. The fix tends to be producing current authoritative sources at consistent cadence.

The completeness of the brand's strongest attributes in the public record. If the strengths aren't written into trusted sources at sufficient density, the engine often can't surface them.

What you can influence but not control directly

Wikipedia. Influence through tier-1 sourcing and disclosed-editor channels. Not direct control.

Sentiment. Heavily shaped by the dominant narratives in tier-1 outlets. You can typically shift it, slowly, with sustained earned media work.

Cross-engine consistency. Each engine pulls from a different mix. Closing inconsistency gaps tends to require authority work across the source types each engine over-weights.

The mental model

Stop thinking about AI reputation as a switch the brand turns on or off. Think about it as a citation graph the brand is constantly editing through earned media, owned content, and Wikipedia work.

The engine's answer tends to be downstream of the graph. Edit the graph and the answer typically changes. Don't edit the graph and the answer becomes whatever the open web produces.

The leverage points, ranked

Highest leverage — Wikipedia. One source, every engine tends to weight it heavily.

Very high — Tier-1 earned media on category-defining prompts.

High — Original research and indices the engines can cite.

Material — Structured owned content as the anti-hallucination floor.

Lower — Mid-tier press, owned blog content, social.

Generally negligible — Most paid content marketing, SEO blog farms, low-authority placements.

A brand that allocates communications spend in roughly this order tends to move the dial. A brand that inverts the order is often paying for vanity.

Where the line really sits

The brands with the strongest AI reputations have not "controlled" what the engines say. They have engineered the inputs the engines retrieve from — over months and quarters, not weeks. The engine then synthesizes the engineered inputs into an answer that tends to reflect the brand the way the brand wants to be reflected.

That isn't control in the absolute sense. It's effective influence — and the most a brand can have short of owning the model itself.