GEO guide
What you receive with the assessment, what we measure before and after, and two anonymised cases: from GEO Score C to B in a quarter, from absence to recommendation in category questions.
An assessment produces two things: a picture (score, sources, risks, plan) and, after the interventions, a measured shift in AI answers. This page explains what you receive, what we compare before and after, and shows two anonymised cases. Names and some details are changed; the numbers are real.
| Indicator | What it measures | Why it matters |
|---|---|---|
| Presence | In how many answers the company appears | The first step: you cannot be recommended if you are not named |
| Accuracy | How many reported facts are current and correct | Being cited with old data can cost more than absence |
| Sentiment | Tone of the answers, per engine | It determines the decision of whoever reads |
| Share of voice | How many competitors are cited next to you, and who | In category questions there are three or four slots |
| Freshness | Median year of cited sources | Dated narratives are born here |
| Covered sources | How many decisive sources are under your control | It shows how much you can act directly |
Context. Italian manufacturing company, 800 employees, difficulty filling technical roles. The GEO Employer Assessment shows a GEO Score C (58/100): the company appears in 42% of answers to branded questions, with mixed sentiment. The heaviest source is a 2021 news item about a reorganisation; the cited reviews have a median age of three years; no source mentions benefits, training or recent awards.
Interventions (6 weeks). Wikidata entry and LinkedIn page updated; a "working with us: facts and figures" page published with dated, verifiable data; replies to the reviews of the last two years and an invitation for recent reviews; an interview with the HR director in a trade publication with data on the training plan; AI crawlers unblocked, stopped by a firewall rule.
90-day re-analysis. GEO Score B (74/100). Presence in answers from 42% to 71%; median source year from 2021 to 2025; the reorganisation still appears, but in one answer in five and alongside the subsequent facts. On Perplexity and ChatGPT the answers cite the facts page and the interview.
Context. B2B service provider, leader in Italy in its niche. In branded questions the answers are correct and positive (score B), but in category questions ("which [service] providers in Italy") the company never appears: the AI cites three competitors, two of them smaller. The source map explains why: the three competitors are listed in two industry directories and in a ranking the AI picks up; the company is in none of the three.
Interventions (4 weeks). Complete listings in the two directories; participation in the ranking with the required data; a category page on the website with definition, comparison and FAQs; Service and FAQPage structured data; two documented replies in a Reddit discussion where the category was being discussed.
90-day re-analysis. The company appears in 60% of category answers on Perplexity and ChatGPT, in 30% on Gemini; in half the cases it is cited first. No change on Claude, which relies more on training memory: expected, and monitored.
We do not promise a position, because answers have no fixed positions, and we do not promise identical timing on every engine: those that browse the web react in weeks, those based on training memory in months. We promise an honest measure: same questions, same engines, before and after, with the sources on the table.
Frequently asked questions
Yes. In industries with many third-party sources (consumer, tech) more interventions are needed to emerge; in those with few sources (niche B2B, industrial) a few well-chosen interventions are enough, but every negative source weighs more.
It is corrected at the source: we identify the page that generates it and act there (update, reply, request for correction). Some engines also allow direct reports; we use them, but the source remains the main lever.
The cases are anonymised by contract; on a call we show real reports with sensitive data blacked out.
Free audit in 48 hours
A free demo audit, no commitment: real AI answers about your brand and an indicative GEO Score, presented on a call.