What to feed LLMs: E-E-A-T in generative search

rich-text

Ask a question in Google today, and chances are you won’t even reach the search results. You’ll get an AI-generated summary first, pulling information from a handful of sources it trusts. Often, that’s the only answer a user needs, and results in no click.

That’s what search looks like these days. SEO isn’t just about chasing the top 10 blue links, but being part of the conversation that helps shape the answer.

And this doesn’t happen by chance; if you want to be part of the answer, you need to feed the system the right signals.

AI selects voices, not just content

In traditional search, you could win by being discoverable. In generative search, you win by being selectable.

Google is explicit that AI Overviews use a customised Gemini model working “in tandem” with its existing Search systems, including quality/ranking systems and the Knowledge Graph. That matters because it means two things are happening at once:

1. Your usual SEO signals still influence what gets pulled in (quality systems, rankings).
2. Entity understanding is baked into selection (Knowledge Graph connections, brand/entity confidence).

LLM chatbots work differently depending on the product, but the outcome feels the same to users: the model curates a small set of sources and turns them into a single narrative.

How AI systems decide what to include

There isn’t one AI selection method, but most systems lean on a combination of:

• Established brand knowledge (what the model already “knows” from training data and long-term signals)
• Fresh web grounding (pulling in current pages to corroborate, update, or cite)
• Entity signals (is this brand/person clearly identifiable and consistently represented?)
• Trust checks (accuracy, safety, reputation, consistency)

This is why newer brands often struggle with AI answers even when their content is good, as the model has less history to work with.

So yes, content matters. But in AI search, reputation and recognition change the starting line.

What E-E-A-T changes in an AI answer world

E-E-A-T is often explained like a content quality framework. That’s true, but in generative search, it behaves more like a trust filter.

1. Experience & Expertise: “Anyone could have written this”

AI systems already have thousands of generic explainers. What they need (and what they reward) is content that signals real-world knowledge:

• First-hand learnings
• Practitioner insight
• Case studies with specifics
• Original analysis and useful viewpoints

If you want to be cited, you need to publish content that’s hard to fake and easy to verify.

2. Authoritativeness: visibility is now a data problem

If a model repeatedly sees your brand mentioned in credible places, alongside relevant topics, it becomes easier to “select” you. This is one reason Knowledge Graph visibility matters so much in Google’s AI experiences.

It’s also why “PR for the sake of it” doesn’t cut through. A random mention isn’t a signal. Consistent, contextual recognition is.

3. Trustworthiness: weak signals won’t cut through

If your content is outdated, inconsistent, anonymous, or unclear about who wrote it, it becomes riskier to cite. And if the model is trying to reduce risk, it will default to sources with clearer trust signals.

Google’s own guidance is aligned with this: be clear about who created your content, how it was created, and why it exists.

How to feed the machines 

Okay, so the title of this blog is slightly misleading… There isn’t a prescribed diet for LLMs. 

The truth is, this will be different from brand to brand. Nuances like the niche you’re in, the subject matter, and how established your brand is will all factor into this. However as a starting point, you should look to:

1. Make authorship unmistakable

• Real bylines
• Credible bios
• Links to profiles and work
• Clear editorial responsibility (who reviews what)

2. Publish content that can’t be spun up in 10 minutes

• Original research (even small-scale is useful)
• Real examples and outcomes
• Expert commentary with an authentic point of view

3. Build mention-worthy presence beyond your site

• Industry publications
• Podcasts/webinars
• Partnerships
• Relevant directories and communities

4. Use the “Who / How / Why” lens on every key page

• Who wrote it?
• How was it produced?
• Why does it exist? 
• What does it help someone do?

5. Treat maintenance as an AI visibility strategy

• Refresh high-performing pages
• Fix outdated claims
• Remove thin content that weakens the domain

An E-E-A-T success story: +43% increase in organic visibility for Lovett Care

media-single-image

rich-text

Following the acquisition of New Care, Lovett Care Group had grown to 28 homes and more than 1,800 beds across England and Wales.

However, Lovett’s digital presence was split across two domains, dividing authority and weakening trust signals. In a high-consideration sector, credibility cannot rely on brand claims alone. Expertise, authoritativeness and trustworthiness needed to be visible and consistent across search.

We consolidated both domains into a single platform, carefully managing migration to protect and transfer authority. We embedded trust signals throughout, bringing together Google reviews, OpenScore ratings, carehome.co.uk feedback and CQC data into a visible review layer. Transparent team bios and content-rich guides demonstrated real expertise, which we strengthened with structured data to make entity relationships clear to search engines.

This translated into a 43% increase in organic visibility, a 56% rise in local pack rankings and a 70% uplift in conversion rate. Read the full case study here.

Will you be skipped or selected?

If AI Overviews and chatbots are your new front page, then your job is to become an easy source to trust.

Ask yourself, could an AI system confidently attribute this answer to you?

If the author isn’t clear, the expertise isn’t provable, the claims aren’t backed up, or your brand isn’t recognised elsewhere, the safe choice is to cite someone else.

So don’t just publish more. Publish content that’s attributable, verifiable, and connected to a real entity people already trust.