How AI assistants decide which brands to recommend

Ask ChatGPT or Perplexity for a recommendation and the assistant names a handful of brands. Those brands usually share one trait. Several sources with no reason to coordinate describe them the same way.

Search Agency has published the method it uses to earn that position for clients. It's called Total Graph Authority. The full whitepaper is free at search.agency/whitepaper.

We built the search.agency website. This post covers what the paper contains and where the build of a site decides how far the method can go.

The five graphs

The method treats brand trust as five graphs. An assistant reads all of them when it picks who to recommend.

• Entity graph. Who the brand is and whether every source agrees on it. The other four graphs build on this one.
• Content graph. The brand's own pages that an assistant quotes from. Clear claims and defined terms make a page easy to lift.
• Link graph. Independent pages that point to the brand and back up what it says about itself.
• Citation graph. Mentions of the brand in sources an assistant already trusts, written in the words a buyer would use.
• Interaction graph. Social and community signals showing that real people use the brand.

When all five agree, the brand holds what the paper calls Total Graph Authority. One version of the brand gets repeated by sources that never compared notes. An assistant reads that agreement as trust.

Eight phases with owners and exit checks

The paper lays out the work as eight phases. Each one has an owner, a set of steps, an output and an exit check.

• 00 Scope and setup. Agree the prompt set, competitors and owners before any work begins.
• 01 Baseline audit. Measure where the brand stands today across every engine and every graph.
• 02 Score and diagnose. A 100-point scorecard turns the audit into a ranked list of fixes.
• 03 Entity fix. Get every source to agree on who the brand is and what it does.
• 04 First-party build. Write the pages an assistant can quote and structure them to be lifted.
• 05 Third-party validation. Earn the independent citations and links that confirm those claims.
• 06 Social utility. Give the community something useful to work with, which makes the interaction graph real.
• 07 Measure and reallocate. Track six outcome KPIs and move effort to the graph with the most to gain.

Every phase includes a worked sample. The paper follows a sample company called Northwind through the whole process, which makes it easier to picture each output on a real brand.

Where the website comes in

Phases 03 and 04 lean hardest on the site. That's the part we see most from the build side.

An entity fix starts on your own domain. Organization schema, the about page and the way the brand names itself in headings all feed the entity graph. When the page copy calls the company one thing and the schema calls it another, an assistant meets two versions of the brand before any third party has weighed in.

The first-party build phase asks for pages an assistant can lift. In practice that means server-rendered HTML and headings that match the questions buyers ask. A page that loads its answer through client-side JavaScript gives a crawler less to quote. We covered the build decisions behind this in a tech stack built to be found.

Who should read it

The paper is written for two readers. Marketing leaders can use it to see what the work involves and who owns each part. A practitioner could pick it up and run the method on their own brand.

Get the Total Graph Authority whitepaper. If you'd rather see where your brand stands first, Search Agency also offers a free AI search audit that scores your brand and named competitors across the engines buyers ask.