Every search agency now has a page about generative engine optimisation. Some of them are doing real work. Some have renamed the content retainer. The difference is visible in about four questions, which are at the end of this piece.
What the two genuinely share
A site that loads quickly, renders its text in the HTML, and is organised sensibly helps in both. Accurate business information helps in both. Writing that answers a real question helps in both. If your search foundations are weak, fix them, and you have done work that counts twice.
Anyone telling you to abandon search to chase assistants is wrong about the numbers. For most service businesses the search results page is still the larger channel by some distance.
Five places they come apart
| Aspect | Search | Assistants |
|---|---|---|
| The unit | A page | A passage |
| The target | A position | A mention |
| The input | Keywords | Questions |
| Trust signal | Links | Agreement across sources |
| Success | A click | Being named, often with no click |
Page versus passage. Search ranks the whole document. An assistant lifts two sentences out of it. A long page can rank first and contain nothing liftable, which is why thorough pages sometimes lose to thinner ones.
Position versus mention. There is no position one in an answer. There is being in it or not, and the same question asked twice may differ. You measure across many questions and over time, not at a point.
Keywords versus questions. An assistant decomposes a vague question into several specific ones before it looks anything up. Pages built around one specific question match those well. Pages built around a service category match none of them well.
Links versus agreement. Links still matter indirectly, because linked pages tend to be found. What decides whether an assistant will state a fact about you is whether independent sources state the same fact. A strong link profile with a contradictory licensing record performs badly here.
Clicks versus mentions. Much of the value arrives with no visit. Somebody asks who to call, hears your name, and searches for you directly or phones. Analytics will book that as direct or branded traffic. Expect the credit to land in the wrong column and plan your reporting accordingly.
The uncomfortable part of this channel is that the win often arrives as a phone call your analytics cannot explain.
What transfers from the work you have already paid for
More than people expect. Technical health transfers almost entirely. Accurate listings transfer entirely. Genuinely useful articles transfer once their openings are rewritten to answer first. Internal linking transfers, because it helps a fetcher move through the site.
What does not transfer is anything built purely to rank: thin location pages, pages written around phrase variants, and link placements on sites nobody reads. These were never good and they are now visibly useless.
Four questions for an agency that says it does GEO
- What is my current baseline, and how did you measure it? A real answer names a set of questions, the assistants tested, and the date. A vague answer means there is no baseline.
- Which of my public records disagree with each other right now? Anyone who has done the work can list them in the first meeting. Anyone who has not will talk about content.
- What will I see in month one? The honest answer is almost nothing visible, because identity work is slow. Promised movement in thirty days is a flag.
- What happens if the assistants change how they cite sources? They do, several times a year. You want a method that rests on being identifiable and useful, not on a trick.
None of this requires you to fire anybody. It does mean asking your current agency to show the baseline before they invoice for the first month of a new line item.