SEO, AEO and GEO
GEO: being recommended by generative engines
Generative Engine Optimization is about how generative engines talk about you: do they recommend you, in what place, in what terms, and from which sources? NeoRank reads it in the same observations as AEO.
What is GEO?
When a user asks “which tool should I choose for…” or “what are the alternatives to…”, the generative engine writes an answer that compares, ranks and recommends. Generative Engine Optimization (GEO) works on that answer: being named among the options, as early as possible, described correctly, and backed by sources the engine trusts.
AEO and GEO overlap widely; the useful distinction is the question. AEO targets the answer to a question (“what is…”, “how do I…”); GEO targets recommendation and comparison. NeoRank measures both with the same tracked prompts and the same observations.
What NeoRank measures for GEO
| Signal | How it is obtained | Where to read it |
|---|---|---|
| Recommendation | Recommending wording in the answer and positive sentiment near your brand | AI visibility (evidence, a prompt's resolution) |
| Position in the answer | The order in which each entity appears; the weighted share counts each mention 1 ÷ position | AI visibility, Competitor analysis |
| Sentiment | A French / English lexicon applied to the excerpts naming your brand | AI visibility's “Sentiment” card |
| Cited sources | The URLs written in answers, classified: your site, a tracked competitor, a third party | Digital PR and citations |
| Competitors' citations | Cited URLs whose domain belongs to an identified competitor | Competitor analysis |
| Traffic from AI | GA4 sessions whose source contains chatgpt, openai, perplexity, gemini, claude, copilot or bing chat | Traffic (“AI traffic share”) |
The “GEO score”
The header's GEO score is your brand's share of voice over the last three observations, shown out of 100: your mentions divided by your mentions plus the tracked competitors'. The computation and its main limit — with no competitor listed in the query set, it is 100 % as soon as you are named — are detailed in share of voice.
weighted share = Σ (1 ÷ your brand's position) ÷ Σ (1 ÷ each entity's position)Sentiment
Each excerpt naming your brand is classified positive, negative or neutral by a lexicon of French and English words. The card shows the share of positive excerpts per engine and up to four quotes. It is a simple, verifiable reading, not a fine analysis: irony or nuance escapes it. With no classified excerpt, the card shows “Not measured”. These are AI answers, not customer reviews.
Traffic from AI
With Google Analytics 4 connected, the Traffic view computes the AI traffic share: sessions whose source contains one of the names above, over all sessions, across the last 28 reported days. It is a floor: an engine that sends no recognisable source (copied link, app without a referrer) does not show up in it.
Google's AI Overviews and AI Mode are not in it as such: their clicks arrive as Google search, and Search Console counts them in the “Web” type with no way to isolate them (Google documentation). For them, the “AI Overviews (Google)” tile shows the automatic reading where it is enabled (observed through a licensed SERP provider, analysed by the NeoRank engine), otherwise the manual reading you record with “Check in Google” — see the AEO page.
Improving GEO
Give the engine something to recommend
- Publish pages per use case and per customer profile: an engine recommends an option for a precise need.
- Keep an honest comparison page (you and the alternatives, explicit criteria): “X or Y?” and “alternatives to X” questions are among the most frequent.
- Expose the verifiable facts that decide: public prices or terms, areas served, lead times, integrations, certifications. A fact missing from your site is one an engine has to guess or take elsewhere.
Consistent facts everywhere
Name, description, offer and contact details must say the same thing on your site, in your JSON-LD and on your external profiles. Inconsistencies end up in answers. Local SEO measures the consistency of the details declared in JSON-LD; AEO covers entity clarity.
Be present in the engines' sources
Open Digital PR and citations: for each third-party domain, the share of each engine's answers citing it. The domains cited without your site are the publications, directories and comparison sites where a mention would bring you into the answer. NeoRank has no media database or outreach tool: the view tells you where to act, not whom to contact.
Cover your market's questions
- Content topics lists the themes derived from your crawled pages; the “GEO opportunity” badge flags a topic that carries questions.
- SEO briefs proposes, from your own pages, a length, an answer format, Schema.org markup and an H2 / H3 outline.
- Track category and comparison questions as prompts, not only your brand name.
Limits to know
- Only Perplexity answers from a web search; the other engines answer from their model, whose knowledge has a date. A page published yesterday can take time to weigh.
- Answers vary from one run to the next: the KPIs cover the last three runs to smooth this, and a sample that is too small is declared insufficient.
- The country is not sent to the engines: it travels in the prompt's text.
- There is no scheduled recurring observation yet: run an analysis or an observation again to refresh the figures.
A four-step method
- Track category and comparison prompts.
- Run an observation and read who is recommended instead of you.
- Fill the missing pages and facts, earn mentions on the domains cited without you.
- Measure again after a new analysis.