GEO - Generative Engine Optimization
More people now get their answer from an assistant instead of a list of links. GEO is the work of making sure your content is retrievable, quotable and correctly attributed when that happens.
What actually changed
For twenty years the transaction was simple. Someone searched, got ten links, and clicked one. Now a growing share of those questions are answered in place: Google shows an AI Overview above the results, ChatGPT and Perplexity return a written answer with a handful of citations, and a portion of users never reach a website at all.
This is not the end of search traffic, and predictions that it is have a poor record. But it changes what visibility means. Appearing as the source behind an answer is now a distinct outcome from appearing as a blue link, and it is possible to have one without the other.
How assistants get their information
There are two mechanisms, and they call for different work.
The first is retrieval. When you ask ChatGPT search, Perplexity or Google's AI Mode a current question, the system runs searches, fetches pages, and writes an answer from what it retrieved. Here your prerequisites are ordinary: your pages must be crawlable, must return their content in HTML, and must be structured so the relevant passage can be lifted out. If a crawler cannot fetch the page, it cannot be cited. That is the entire mechanism, and it is why most GEO work starts as technical work.
The second is training. Models absorb a large corpus during training, and what they will say about your company with no retrieval at all reflects what was written about you across the web, not what is on your homepage. You cannot edit that directly. You influence it the slow way, by being described accurately and consistently in the places that get indexed widely: your own documentation, industry publications, directories, comparison sites and communities where your category is discussed.
The parts you can control
The controllable surface is smaller than most GEO pitches suggest, and it is worth being specific about it.
Crawler access. GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended are distinct user agents with distinct effects, and blanket rules in robots.txt or a CDN bot filter frequently block them by accident. This is the single most common finding we make.
Extractability. A model quotes what it can isolate. Facts stated in a sentence get used; the same fact implied by an infographic does not. Direct answers near the top of a section survive summarisation better than answers built over five paragraphs.
Entity clarity. Consistent company naming, structured data, and a page that plainly states what you do and where you operate. Models conflate similarly named organisations, and the fix is to make the connection unambiguous.
Presence in the sources models reach for. In many categories the assistant's answer is assembled from third-party comparisons and industry sites rather than vendor pages. If you are absent from those, your own site can be perfect and still not appear.
Measurement is different, and worse
There is no rank tracker for this, and you should be suspicious of tools that present one as though there were.
Assistant answers are generated, not retrieved from a ranked list. Ask the same question twice and you may get different sources. Answers vary by account, by region and by model version, and providers change the systems without notice. Anyone showing you a precise position for "your visibility in ChatGPT" has invented a metric.
What works is sampling. We define a prompt set that reflects how your buyers actually ask, run it across several assistants on a fixed schedule, and record whether you appear, in what role, and how you are described. Alongside that, we track referral traffic from assistant domains in analytics, which is genuinely measurable but systematically undercounts, because the reader who got their answer without clicking never shows up.
Both signals are noisy. Together they show direction over months, which is enough to decide whether the work is worth continuing.
A reasonable expectation
GEO is early. The mechanisms are partly undocumented, they shift, and the honest version of this service is an ongoing experiment run with discipline rather than a playbook with guaranteed outcomes.
What makes it worth doing now is that most of the work is not speculative. Crawlable pages, clear structure, accurate entity data and content that states its facts plainly are all things that help conventional search too. The overlap means you are not betting the budget on one prediction about how people will search in three years.
What you get
AI visibility baseline
We run a defined set of prompts across the major assistants and record what they say about you, your competitors and your category. This is the starting point everything else is measured against.
Crawler access review
GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended are separate agents with separate consequences. We check what your robots.txt and CDN actually allow, because these are often blocked by accident.
Retrievable page structure
Content restructured so a specific answer can be extracted from it: clear headings, self-contained sections, definitions and figures stated in text rather than implied by a graphic.
Entity and attribution work
Consistent naming, structured data and an about-us surface that lets a model connect the page to your organisation rather than to a similarly named company.
Source and mention strategy
Assistants draw on more than your own site. We identify the industry sources, directories and comparison pages that models tend to reach for in your category, and work on how you appear there.
Recurring measurement
Prompt sets re-run on a schedule, plus referral tracking from assistant domains in your analytics. The point is a trend line, not a single flattering screenshot.
How we work
- 01
Establish what assistants currently say
We assemble prompts that match how your buyers actually ask, run them across several assistants, and log the answers and cited sources. Answers vary between runs, so we sample rather than test once.
- 02
Find out why
Missing citations usually trace back to something concrete: a blocked crawler, content that only exists inside JavaScript or a PDF, or a category where third-party sources dominate and you are absent from all of them.
- 03
Fix access and structure first
Nothing else matters if the content cannot be fetched or parsed. This overlaps heavily with technical SEO, which is why we usually run the two together.
- 04
Rewrite for extractability
Pages get the direct answer near the top, specific facts stated plainly, and sections that make sense when lifted out of context. Which is exactly how a model will use them.
- 05
Re-measure and adjust
We repeat the prompt set at a fixed interval and compare. Changes in AI answers are slower and noisier than ranking changes, so we look at direction over months rather than week to week.
Tools and technology
Where a solid open-source tool exists, we choose it over a closed one. No lock-in to a single vendor, and costs you can actually predict.
- Schema.org
- Ollama
- Playwright
- Matomo
- Umami
- ChatGPT
- Anthropic Claude
- Perplexity
- Google Gemini
- Google AI Overviews
- Microsoft Copilot
- Google Search Console
Frequently asked questions
More services in this category
Technical SEO
Search engines cannot rank what they cannot crawl, render or index. Technical SEO removes the obstacles sitting between your content and the results page.
Content SEO
Content earns search visibility when it answers a question someone is actually asking. We find those questions, then write and optimise the pages that answer them properly.
SEO analysis, monitoring and optimization
SEO and GEO are ongoing processes, not one-off projects. We monitor what is working, explain what changed when results move, and turn that into the next set of decisions.
Read testimonials from companies that trusted us
Delivered well ahead of the deadline.
ZanReal's individual approach is impressive.
Knowledge and business intuition make them a valuable partner.
Quick solutions that reduced costs by 99%.
Latest
Technical guides, security insights, and what we've learned building with AI.
Do you know what AI assistants say about your brand?
Message usTell us which questions your buyers put to assistants, and we will run a baseline that shows how often you are cited and where to start.
Can't keep up with changes in AI world?
Let us do the heavy lifting. Every week we distill the most important AI developments into a focused 5-minute briefing - so you stay ahead without the noise.
Find out more