Credible Roots / Generative Engine Optimization
Generative engine optimization for companies
GEO is an entity problem, not a content problem
Generative engine optimization is the work of making a model describe your company accurately and favourably when it generates an answer about your category. That is decided by what the web collectively says about you, not by what you publish about yourself — which is why most GEO retainers, which are content retainers, do not move it.
One named client engagement, published with permission, and one figure we would rather you knew before buying. The case study · the research, with sources.
Generative engine optimization is the practice of shaping how large language models describe an entity when they generate answers about its category. It differs from answer engine optimization in what it acts on: AEO makes a specific passage retrievable and quotable at the moment a question is asked, while GEO works on the model’s underlying picture of who you are, which is assembled from the whole web rather than from your site. That makes three things decisive. First, entity resolution: consistent structured data, one canonical identity, matching details across every profile, and a Wikidata record where the company genuinely qualifies, so the model is not merging you with three similarly named firms. Second, consensus: models reproduce what many independent sources agree on, so a claim stated one way on your site and five other ways elsewhere resolves to the version the web repeats, not yours. Third, access, because an engine that cannot fetch your pages cannot ground an answer in them. The honest constraint is that GEO is the slowest of the three disciplines and the least controllable, since most of the material a model learns from is not yours to edit.
- GEO acts on the entity. AEO acts on the passage. Same floor, different jobs.
- Consensus beats your own copy. What the web repeats is what the model says.
- Ambiguous identity is the most common fault, and structured data fixes it.
- It is the slowest of the three. Anyone quoting you a date is guessing.
On this page
What generative engine optimization actually is
When a model answers “who are the leading firms in X”, it is not looking anything up in the sense a search engine does. It is producing the most probable continuation given everything it has absorbed about X, sometimes grounded by a live retrieval step. Your company appears in that answer, accurately or otherwise, to the extent that the sources feeding it agree you belong there and agree on what you do.
This is why GEO resists the tactics that work elsewhere. You cannot keyword your way into a consensus. You can, however, change three things that reliably feed one: whether you are identifiable as a single entity, whether independent sources describe you consistently, and whether the engines can reach your material at all.
GEO vs AEO vs SEO
The terms are used interchangeably, including by people selling them. They are not the same job.
| SEO | AEO | GEO | |
|---|---|---|---|
| Acts on | A page | A passage | An entity |
| Question it answers | Does this page rank for the query | Is this passage worth quoting right now | Does the model know who we are, and say so accurately |
| Main lever | Relevance and links | Extractability and crawler access | Entity resolution and third-party consensus |
| How much is on your own site | Most | Some | Least |
| Speed | Months | Weeks for access, months for position | Slowest |
| Failure mode | Outranked | Retrieved but not quotable | Confused with another company, or absent entirely |
We run all three as one engagement because they share a floor. A site that cannot be crawled, rendered or resolved fails at all three at once, and that is the most common situation we are called into.
What we build
An entity audit
What the major models currently say about you, recorded against a fixed prompt set. The frequent findings are unflattering and useful: the model has merged you with a similarly named company, attributes your work to a competitor, has you in the wrong country, or has never heard of you. Each of those has a different fix.
Resolution
Organization and Person structured data describing something real, one canonical identity page, a sameAs graph tying your profiles together, matching name, address and description everywhere you appear, and a Wikidata item where the company genuinely qualifies. This is the part that converts a string into an entity.
Consensus work
Getting independent sources to state the same facts the same way. Directory and database records corrected. Trade coverage pitched. The specific, checkable claims you want repeated written so they are easy to repeat. This is the slow half and the half that actually moves a model’s description of you.
Access
Crawler reachability confirmed per engine from server logs rather than inferred from robots.txt, plus pages that exist without JavaScript. Which crawlers to allow is a decision with trade-offs.
A baseline you can re-run
The prompt set, frozen and versioned, re-run on a schedule with the sample size reported. The metrics we use are defined publicly so you can reproduce or dispute them.
How GEO is measured
Against a fixed prompt set, repeatedly, with the sample size attached — not against a rank, because no rank exists. The measurements are presence rate, citation share, the substitution set of competitors named when you are not, crawler reachability per engine, and corroboration count.
The one specific to GEO is description accuracy: of the answers that do mention you, how many describe you correctly. A company can have a healthy presence rate and a serious problem, if half those mentions place it in the wrong category.
What nobody can promise
- That a model will describe you the way you would write it. Most of the material it learned from is not yours to edit.
- A position, score or ranking. None exists in a generated answer.
- Removal of a competitor from an answer. Not a lever anyone has.
- A date. GEO is the slowest of the three disciplines and depends on third parties.
- Correction of a specific model output on request. You can change the evidence; you cannot edit the answer.
Who this is for, and who it is not
| A good fit | Not yet | |
|---|---|---|
| Identity | Ambiguous or confused entity, or a rebrand the web has not caught up with | Already unambiguous, well covered and correctly described |
| Category | Buyers ask models for shortlists and comparisons | Purely transactional or walk-in demand |
| Coverage | Has independent coverage, or will invest in earning it | Expects on-site changes alone to be enough |
| Horizon | Buying position over a year or more | Needs attributable pipeline this quarter |
Questions companies actually ask
What is the difference between GEO and AEO?
AEO makes a specific passage retrievable and quotable at the moment someone asks a question, so it acts on your pages. GEO works on the model’s underlying description of your company, which is assembled from the whole web, so it acts on your entity and on what third parties say. AEO moves faster and is more within your control; GEO is slower and matters more for whether you are considered at all.
Is GEO just SEO with a new name?
Partly, and we say so. It shares the technical floor with SEO and a great deal of the work is ordinary good practice. What is genuinely different is that entity resolution becomes a deliverable rather than a nice-to-have, third-party consensus outweighs your own copy, and measurement is against a prompt set rather than a rank tracker.
The model says something wrong about us. Can you fix it?
Not directly, and nobody can. There is no field to edit. What can be done is change the evidence the description is drawn from: correct the database and directory records, resolve the entity properly with structured data and Wikidata, and get independent sources to state the accurate version. Then re-measure. It is indirect and it is slow, and any vendor promising a direct correction is describing something that does not exist.
How long does GEO take?
Longer than the other two, and we will not give you a date. Access faults resolve on a crawler’s schedule, often within weeks. Entity resolution shows up as models refresh. Consensus depends on third parties publishing, which is not ours to schedule.
Do we need a Wikidata item?
Only if the company genuinely meets the inclusion criteria, which are lower than Wikipedia’s but real. Where it qualifies it is one of the more direct entity signals available. Where it does not, creating one invites deletion and wastes the effort. The criteria are set out here.
What does an engagement cost?
$2,000 a month, three month minimum, then month to month. The same single price covers the answer-engine work, the entity work and the founder track — no tiers and no setup fee. The pricing page sets out what it does not buy, and why we sit below the range circulating for this category.
Related reading
- Answer engine optimization: the retrieval half
- The metrics, with formulas and sample sizes
- Wikidata: the inclusion criteria, honestly
- Knowledge panels, and the five things that move them
- Which AI crawlers should you allow
- What published method looks like on a client site
- The agency field, assessed on what it publishes
- The statistics, traced to source
- GEO for law firms, inside Rule 7.1
Sources
How this was made: written from client work, with crawler and structured-data details taken from each vendor’s own documentation rather than from secondary summaries. Drafted with AI assistance, then reviewed and approved by the named author before publication. Our editorial standards.
Find out what the models currently say about you
The entity audit is the first thing we run, and it is usually the uncomfortable one. We will show you how the major models describe your company today, who they name instead, and whether they can reach your site at all.
Book a 30 minute call