Become the brand recommends.

Search optimisation earns a link in a list of ten; generative engine optimisation earns a sentence inside a single answer. The underlying work overlaps almost entirely, because crawl access, structured data and authority decide both. What differs is the target: a model does not rank pages, it assembles a response from the sources it can parse and has reason to trust.

So the two are run together rather than chosen between. The last mile is where they part company, and it is mostly about how a passage is written and whether the business can be identified as an entity.

Timelines split by what is actually broken. Technical fixes register quickly: once a model can crawl and parse your pages, the change can surface in answers within weeks, because access is a switch rather than a reputation. Authority takes a quarter or more, since the sources an assistant leans on have to exist and be cited before they count for anything.

Anyone quoting a fixed number across both is quoting a sales figure. The honest answer depends on which of the two is holding you back, which is what the audit establishes first.

Replacing your search agency is usually the wrong move. Traditional search still sends real traffic, and the foundations it maintains, crawl access, rendering and structured data, are the same foundations generative visibility is built on. Tearing that out to start again costs you the groundwork and buys nothing.

We work alongside an existing team, or take the technical layer off their hands if that is simpler. We will not duplicate work someone is already doing well.

KoldOS does the development, and that is the part worth paying for. Most generative engine optimisation engagements end with a document your engineers have to action, and it queues behind the product roadmap until it is stale. We write the schema, change the templates, fix the rendering and ship it.

If you would rather your own team implement, we hand over specs and review what lands. The deliverable is the change, not the recommendation.

A rebuild is usually unnecessary. Most sites need their markup, rendering and information architecture fixed rather than replaced, and that work happens inside the platform you already run. Rebuilding is a way of paying for the fix twice when the existing site would have carried it.

We recommend rebuilding only when the platform itself blocks the work, and we say so before you have spent anything on it.

The assistants we optimise for are ChatGPT, Google's AI Overviews and Gemini, Perplexity, Claude and Microsoft Copilot. Each fetches pages using its own agents and draws on an overlapping but different set of sources, so being named by one says very little about the others.

Each is therefore tracked separately. A single blended visibility score hides which surface is losing, which is the only thing the number was meant to tell you.

Measurement is share of answer rather than rank. A fixed set of the questions your buyers actually type is put to each assistant on a schedule, and we record how often you are named, which sources were cited when you were, and who appeared instead when you were not. Model answers vary between runs, so the signal is the trend rather than any single check.

The prompts are published alongside the numbers. A share-of-answer figure reported without its questions cannot be reproduced or argued with, which makes it a claim rather than a measurement.