Step 01 · Weeks 1 to 3
Diagnose
Establish the baseline. Investigate buyer questions, AI answers, competitors, sources, entities, claims, technical foundations and business measurement.
Service 01
Get understood, cited and recommended in AI answers
Generative Engine Optimization Services
Updated · by Underneath team
Generative engine optimization is the work of making a business easier for AI search and answer systems to understand, retrieve, describe and reference when buyers are deciding what to buy, who to trust or where to go. It is not the same thing as writing content for AI.
We investigate the information underneath the answer: the questions buyers ask, the entities involved, the claims being made, the sources behind them, where that information is missing or contradicts itself, and what the competitors appearing instead have that you don’t. Then we change the parts that are holding you back and measure whether the change worked.
Diagnose → Architect → Deploy → Compound. That’s the Underneath Method.
Underneath runs GEO programs for local and multi-location businesses, ecommerce, SaaS and enterprise brands. Depending on the diagnosis, the work covers AI visibility analysis, content and information architecture, technical GEO, entity and source work, measurement and team enablement, on WordPress, Shopify, headless or custom stacks.
The answer is only the surface.
A buyer asks:“What’s the best accountant for a small business?”
Answer
“…three firms worth considering are Your firm[1], Firm B[2] and Firm C[3]…”
Underneath that answer:
- Which firms does the system understand as accountants?
- Which firms are associated with small businesses?
- Which claims about each firm can be supported?
- Which sources describe them?
- Which information is current?
- Where do sources agree or disagree?
- Why does one competitor appear while another does not?
That’s where our work starts: we work backwards from important answers to understand the information environment underneath them.
What is generative engine optimization?
Make the right information available to the systems that shape the answer.
Generative engine optimization (GEO) is the practice of improving the information, technical foundation and external presence around a business so that AI-powered search and answer systems can discover, understand and reference it appropriately.
The work builds on SEO. Crawlability, useful content and clear information architecture all still matter, and structured data can still help systems understand what a page represents.
GEO adds a further question: given everything published about this business, on its own site and elsewhere, what will an answer system understand, retrieve and say? That is the layer we investigate.
Google’s guidance on AI features (opens in a new tab) says SEO best practices still apply to AI Overviews and AI Mode, and that no special optimizations are needed to appear in them. So we don’t treat GEO as a replacement for SEO. We treat it as a broader information problem that sits on top of a sound technical and search foundation.
Why GEO matters
A buyer can now get the shortlist before visiting a single website.
A search results page gives a buyer ten links to investigate. An AI answer may give them a short list of options, already described and compared, with a handful of sources. That changes the visibility problem. Ranking for the question is no longer enough. The information around your business has to be strong enough that, when the question becomes an answer, you are still in it.
Our research looks at how AI answers cite sources, how often assistants agree and how consistently recommendations hold up when the same question is asked again. The underlying point: AI answers are not static rankings. They vary by question, platform, source set and time.
So we measure patterns, and never treat one screenshot as a result.
We reverse-engineer the answer without pretending we can see inside the model.
For important customer questions, we work backwards from observable outputs.
- The question. What are buyers actually asking?
- The answer. Who appears, how are they described, and what is being recommended or compared?
- The evidence. Which claims and sources support the answer?
- The information system. How are the business, products, people, services and other entities represented across the web?
- The gap. What is missing, inconsistent, outdated or stronger for a competitor?
- The intervention. What should change?
- The measurement. What would demonstrate that the change mattered?
We don’t publish every detail of how we do this. What a client needs is not the recipe but a stated reason for every change we make.
GEO starts with diagnosis, not a content calendar.
The exact work depends on what the diagnosis finds. A typical engagement can include the following.
AI visibility and answer analysis
We establish how the business currently appears in AI answers and search results for the questions that drive buying decisions.
We examine more than whether the brand is mentioned. We look at descriptions, recommendations, competing entities, citations, source patterns and how those observations change across related questions.
Deliverable: a baseline and visibility map tied to your priority buyer questions.
Query and buyer-intent research
We identify the questions behind the buying journey: category, comparison, problem, product, location and high-intent questions. We then examine how related formulations change the answer.
We are not trying to collect thousands of prompts. We are looking for the question families that expose real visibility gaps.
Deliverable: a prioritized buyer-question set.
Answer and claim analysis
We decompose important answers into the claims being made about the companies, products or services involved. Then we investigate the information supporting those claims. We look for:
- Important claims
- Supporting sources
- Missing evidence
- Conflicting information
- Outdated information
- Competitor advantages
- First-party vs third-party descriptions
Deliverable: a prioritized map of information gaps and opportunities.
Content and information architecture
We improve the pages the diagnosis points to, which can mean rewriting, restructuring, consolidating or, where something is genuinely missing, creating new pages. The aim is clearer information about the questions, entities and claims buyers care about, not more content. That can mean:
- Stronger definitions
- Clearer explanations
- Question-to-answer structure
- Internal relationships between topics
- Useful comparison information
- Better evidence
- Less duplication
Deliverable: priority pages changed for a defined reason.
Technical GEO
Important information has to be reachable before it can be understood. We investigate the technical conditions that get in the way. Depending on the site, this can include:
- Crawl access
- Rendering
- Indexability
- Canonicalization
- Internal linking
- URL architecture
- Structured data
- JavaScript-dependent content
- Redirects and status codes
- Robots directives
- Bot and security controls
One example: OpenAI documents OAI-SearchBot (opens in a new tab) as the crawler behind ChatGPT search and recommends allowing it in robots.txt for sites that want to appear there. A security rule that blocks it can quietly remove a site from those answers. That is why we handle technical work as part of the visibility problem, not as a checklist of isolated SEO tasks.
Entity understanding
We investigate how the business is represented as an entity across its own properties and relevant external sources. The work can include:
- Organization and brand relationships
- Products and services
- People and experts
- Locations
- Important attributes
- Consistency across sources
- Structured relationships
- Author and expertise signals
- External entity references
We are not trying to manufacture a “knowledge graph.” We are making the basic facts about the business clearer and more consistent wherever they appear.
Source and evidence strategy
AI answers draw on what the rest of the web says about a company, not just its own site. We identify the sources behind the questions and claims we’re investigating, then work out where the business has:
- Strong first-party evidence
- Independent corroboration
- Missing evidence
- Conflicting information
- Outdated information
- Competitor gaps
Where it is warranted, we help strengthen that footprint through legitimate source development: accurate listings, useful material other sites have reason to reference, and corrections where published information is wrong. We don’t manufacture citations.
Monitoring and experimentation
We track the priority questions continuously, but we don’t treat every movement as proof. We set a baseline, form a hypothesis, make the change and retest the relevant questions.
Every finding is reported as observed (what the data directly shows), correlated (what moved alongside something else), attributed (what the available tracking connects to a business outcome) or hypothesized (a likely explanation we still need to test).
AI systems change often. Without those labels, a report can’t tell a result from a coincidence.
Team enablement
We don’t want your team to depend on an agency forever. We document the principles behind the work and give your content, marketing and development teams what they need to keep it up. Depending on the engagement, that can include:
- Content guidelines
- Information architecture principles
- Technical checklists
- Entity and source guidelines
- Measurement routines
- Team workshops
- Review processes
We work across the answer experiences your buyers actually use.
Every program tracks ChatGPT, Claude, Gemini, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot and Bing. Which of them matter most depends on your market and how your customers research.
We don’t assume the platforms behave alike. We measure each one separately and look for the patterns that hold across them, so the program is never built around one platform’s behavior this month.
Visibility is not the same thing as business impact.
Our measurement model has four layers.
| Layer | The question | What we look at |
|---|---|---|
| 01 · Visibility | What is being said? | Brand presence, recommendations, descriptions, competitive presence, citations, query coverage |
| 02 · Evidence | What supports it? | Claims, sources, corroboration, conflicts, freshness, information coverage |
| 03 · Behavior | What did people do? | AI referrals, engagement, calls, forms, applications, bookings, conversions |
| 04 · Business | Did it matter? | Qualified leads, pipeline, customers, revenue, acquisition cost, other agreed business outcomes |
We agree the relevant metrics before the engagement begins. We do not collapse everything into a single “AI score.” Every metric and how it is reported is set out in how we measure.
Different surfaces. Overlapping foundations.
SEO, AEO and GEO share most of their technical and content fundamentals. What differs is the outcome being measured. The terms are defined in the AI search glossary, and the agency-level comparison is in GEO agency vs SEO agency.
| Column 1 | SEO | AEO | GEO |
|---|---|---|---|
| Primary outcome | Search visibility and clicks | Direct answers | AI-generated answers and recommendations |
| Typical surface | Search results | Featured snippets and direct answers | AI search and answer experiences |
| What matters | Pages, queries, links, technical foundation | Clear answers and structured information | Information, entities, evidence, sources and answer visibility |
| Measurement | Rankings, impressions, clicks, conversions | Answer visibility | Mentions, recommendations, citations, referrals and business outcomes |
A technically weak, poorly structured site is still a problem, and GEO won’t paper over it. But a technically excellent site can still have an information footprint that doesn’t produce the answers the business wants. That’s where GEO begins.
Four stages. The last one doesn’t end.
Step 02 · Weeks 2 to 4
Architect
Turn the findings into a prioritized plan. What needs to change? Where? Why? Who owns it? How will we know?
Step 03 · Month 2 onward
Deploy
Ship the work, on the website and in the sources beyond it.
- Technical
- Content
- Entity
- Source
- Evidence
- Measurement
Step 04 · Ongoing
Compound
Measure what the changes did, test the hypotheses and set the priorities for the next cycle. When the evidence changes, the plan changes with it.
From signature to shipped work, without a long silence.
| When | Milestone | What happens |
|---|---|---|
| Day 0 | Agreement and access | Goals, markets, competitors, measurement and access agreed. |
| Days 1 to 7 | Audit delivered | Baseline across AI visibility, search, the website, sources, entities, customer language and measurement. |
| By day 30 | Plan agreed | Priorities, pages, sources, interventions, owners and measures. |
| Days 31 to 60 | First work live | Priority changes ship, with the baseline kept so they can be measured. |
| Day 90 | First full review | What changed, what didn’t, what we learned and what happens next. |
GEO matters most when buyers are already asking questions.
We work with:
- Local and multi-location businesses, where buyers ask AI where to go, who to call or which provider to choose.
- Ecommerce, where buyers ask what to buy, which product fits their needs or how products compare.
- SaaS, where “best X for Y,” alternatives and category questions influence the shortlist.
- Enterprise brands, where long buying cycles, multiple stakeholders and a large information footprint make consistency and discoverability important.
What they share: customers research the decision before they contact you. If you are deciding whether to hire help, start with Is a GEO agency worth it? and How to choose a GEO agency.
Scope determines the program.
Underneath’s GEO programs are priced in three bands: $5,000 to $10,000, $11,000 to $20,000, or $20,000+ a month. The band depends on the scope and complexity of the work, not on company size alone.
Every program begins with the Marketing & AI Visibility Audit. The audit can also be bought on its own, and if you continue into a program, its fee is credited toward it. Fixed-scope projects are priced to scope.
Questions about GEO
What is generative engine optimization?
It is the work of making sure AI assistants and AI search can find accurate information about your business, understand it and cite it when buyers ask the questions you should be answering: on your own site, in its technical setup and in the sources elsewhere that answers draw on.
Is GEO just SEO for AI?
No. GEO depends on SEO fundamentals, but the problem is broader. Beyond whether a page can rank, we look at how the business is represented across questions, sources, claims, entities and answer experiences.
Can you guarantee that an AI assistant will recommend us?
No. AI systems are dynamic and their answers can vary by question, user context, source availability and platform. We can establish a baseline, improve the information system around the business, measure observable changes and report them honestly.
Do you create lots of AI-generated content?
No. Volume is not the goal. We create, restructure or consolidate pages where the diagnosis shows a need, and often the answer is fewer, clearer pages.
Do you build citations?
Not in the link-building sense. We find the sources that answers draw on for your questions, correct what is wrong there and help create useful, credible information that others have reason to reference.
How long does GEO take?
There is no universal timeline. Changes to your own pages can show up once they are recrawled; work on external sources usually takes longer. We set the baseline first, and the first 90 days are about getting measurable work live against it.
How do you measure results?
In four layers: visibility, evidence, behavior and business. The exact metrics are agreed at the start of each engagement, and how we measure explains each one.
Can you work with our existing SEO agency?
Yes. GEO usually sits alongside SEO, content, PR, development and in-house marketing. We work with the partners you already have rather than replacing them.
Can you work with any CMS?
Yes. We work on WordPress, Shopify, headless and custom stacks. The implementation changes with the technology; the method doesn’t.
What happens when AI platforms change?
We keep tracking the same questions and sources, so a platform change shows up in the data. When it does, we investigate what changed and adjust the work. Because the strategy isn’t built on one platform’s current behavior, a change on one platform rarely means starting over.
Free strategy call
Start with the questions your customers are already asking.
On a free 30-minute strategy call, we look at the questions your buyers ask and then send a short written read: what we see, what we would investigate first, and whether we’re the right fit.
We don’t sell the audit on the call.