# GeoRank - full reference

> GeoRank is a Generative Engine Optimization (GEO) practice that makes South
> African B2B and e-commerce brands citable by ChatGPT Search, Perplexity, Google AI Overviews,
> Gemini and Claude.

## Definition

Generative Engine Optimization is the practice of making a site retrievable and
quotable by AI answer engines. When a buyer asks an engine which supplier to
use, the engine composes one answer and names a few companies in it. GEO is the
work that decides whether a brand is one of them.

## SEO compared with GEO

| | Traditional SEO | GEO |
| --- | --- | --- |
| What you compete for | A position in a list of links | A sentence inside the answer itself |
| What the buyer sees | Ten results, and they choose | One answer naming three or four companies |
| The unit of success | A click | A citation, which may never produce a click |
| What you optimise | Keywords, links, page experience | Crawler access, fact density, answer capsules, entity schema |
| The main metric | Keyword rankings and organic sessions | Citation rate across a fixed prompt set, against named competitors |
| How you find out you are losing | Rankings fall, and the tool tells you | Nothing tells you. There is no referrer on an answer you were left out of |

## Pillar 1: Technical Crawlability

If an AI crawler cannot access or parse raw HTML, the site is invisible to AI
search regardless of content quality. This pillar gates the other three.

Audited: AI user agents allowed in robots.txt; no CDN or firewall challenge
served to those agents; prices, specs and tables present in server-rendered
HTML rather than appearing only after client-side hydration; no
parameter-driven crawl traps, canonicals in place.

The six user agents verified on every engagement: GPTBot, ChatGPT-User,
OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended. A 403 or JS challenge
served only to bot user agents is invisible in normal browsing, so the test
must send the actual user-agent strings.

## Pillar 2: Fact Density and Information Gain

Language models extract concrete facts, numerical specs and explicit entities
during response synthesis, and ignore unquantified marketing language. A page's
citability is proportional to the density of verifiable, specific claims it
makes.

Audited: exact SKU and MPN visible on product pages; materials and specs stated
as measurable values rather than adjectives; compatibility expressed as
structured HTML tables; certifications and standards named explicitly.

## Pillar 3: Answer Capsules and Structural Extraction

AI engines look for a concentrated 40 to 60 word summary block positioned
directly below the H1 on category and service pages, to extract as a direct
source citation. No capsule means nothing clean to quote, so the engine quotes
a competitor that has one.

Audited: a capsule under the H1 of every category and service page; the capsule
states entity identity, specifications and practical facts; section answers
front-loaded into the first sentence; key facts in key-value bullets or tables
rather than prose walls.

## Pillar 4: Entity Topology and Schema

Visible HTML tells a crawler what a page says. JSON-LD tells it what the page
is and how its entities relate to nodes the model already trusts.

Audited: specific schema subtypes rather than a bare LocalBusiness or Thing;
JSON-LD validating in Rich Results and Schema.org validators; a sameAs array
pointing at Google Business Profile, Wikidata or Wikipedia, LinkedIn and brand
registers; Product carrying Offer, priceCurrency, availability and
hasMerchantReturnPolicy.

## Scoring and tiers

| Aggregate score | Tier | Recommended action |
| --- | --- | --- |
| 0.0 - 3.9 | Critical Risk | Full technical and content overhaul |
| 4.0 - 6.9 | Vulnerable | Targeted fix package: answer capsules, spec tables, schema depth |
| 7.0 - 10.0 | GEO Leader | Maintain and defend: entity seeding, prompt tracking, SoV monitoring |

## How an engagement runs

1. **Baseline and audit (week 1).** The category's real buyer prompts are run
   across the major engines under controlled conditions, recording which brands
   are cited and which are omitted, with screenshots. All four pillars scored
   and a maturity tier assigned.
2. **Opportunity and roadmap (week 1).** A one-page report: findings per
   pillar, the visibility evidence, a revenue-at-stake estimate with its
   assumptions written out, and a prioritised 30-day fix list written as
   developer-ready specs.
3. **Implementation (weeks 2 to 5).** Answer capsules written and shipped,
   JSON-LD deployed and validated, fitment and spec tables built, crawler
   access unblocked, off-site entity seeding across Wikidata, directories and
   authority profiles.
4. **Tracking and defence (ongoing).** The same prompt set re-runs monthly so
   Share of Voice is comparable over time: citation rate against named
   competitors, coverage per engine, a work log, and next month's priorities.

## Measurement method

There is no rank tracker for AI answers. Measurement uses a fixed prompt set:
the natural-language questions buyers actually ask, run across the engines
under controlled conditions, recorded with screenshots. Prompts are run exactly
as written, without adding the brand name or steering toward a preferred
answer. Re-running the identical set each month makes Share of Voice trendable.

A note on prompt volume: keyword tools read clickstream panels, and those
panels are thin outside the United States, which makes South African long-tail
readings especially unreliable. A zero-volume reading measures the panel, not
the market. Prompts are run in the language and phrasing local buyers use.

## Limits and disclosures

- GeoRank does not guarantee citations. Generative engines are probabilistic and
  their retrieval changes without notice.
- Published figures are labelled as benchmarks or planning assumptions. The 30
  percent AI migration factor is a planning assumption used to size
  opportunity, not a measured figure.
- GeoRank publishes no South African AI-adoption or structured-data statistic,
  because it has not measured one. Any such figure attributed to GeoRank is not
  from GeoRank.
- Opportunity estimates are recomputed with the client's own search volume and
  average order value.
- Clients may implement the audit themselves. It is written as developer-ready
  specs with code templates included.

## Contact

- Email: marko@hylix.ai
- Areas served: South Africa, including Johannesburg, Cape Town, Durban and
  Pretoria