Generative engines — ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews — are now a primary channel through which prospective customers, journalists, investors and graduates form a first impression of a company. We ran 240 prompts (20 per company) across four engines for 12 leading Moroccan brands to measure who gets cited, who gets described negatively, and who simply does not exist in the answer.
Methodology in brief
We selected 12 companies spanning banking, telecoms, mining, agri-food and retail. For each, we drafted 20 prompts across four intent types: factual ('what is [company]'), evaluative ('is [company] reliable'), comparative ('[company] vs [competitor]'), and recommendation ('best [category] in Morocco'). Each prompt was run on ChatGPT (GPT-5), Perplexity (Pro), Gemini (2.5 Pro) and Claude (Sonnet 4.5) in fresh sessions in January 2026. We coded every response for citation presence, sentiment, and factual accuracy against a verified baseline.
Perplexity leads on citation rate, which is expected — its architecture surfaces sources explicitly. The more consequential finding is the gap between Gemini (52%) and Claude (41%): two engines that shape investor and enterprise audiences respectively are significantly under-citing Moroccan brands. Claude's low rate reflects its weaker web-retrieval grounding; Gemini's reflects a thinner Moroccan training surface.
The three quiet winners
Three companies outperformed the cohort on a composite visibility score combining citation rate, sentiment, and factual accuracy: OCP Group, Maroc Telecom, and Attijariwafa Bank. They share three properties: dense, structured, regularly updated English and French Wikipedia presence; a high volume of indexed press releases with quantified facts; and consistent entity disambiguation (the brand name resolves cleanly to the company, not a homonym).
Top 10 prompts by citation divergence
The prompts below produced the widest divergence between engines — the cases where a brand is highly visible on one engine and invisible on another. These are the highest-leverage gaps to close.
Sentiment when cited
Being cited is necessary but not sufficient. Of the citations we recorded, 71% were neutral or positive, 19% mixed, and 10% clearly negative. The negative citations cluster around three topics: telecoms customer service, water-utility service quality, and historical mining incidents. Critically, negative citations persist longer than positive ones — the same 2019 incident reappears in 2026 answers because no fresher, higher-authority source has displaced it.
What drives visibility
Regression analysis across our 12 companies shows three factors explain 68% of citation-rate variance: structured-data richness on the corporate domain (schema.org Organisation, FAQ, NewsArticle); the recency and volume of indexed press coverage in French and English; and the presence of a well-maintained Wikipedia article with at least ten independent secondary sources. Brands missing the Wikipedia article underperform by an average of 23 percentage points.
“We used to optimise for Google's first page. We now optimise for the first sentence of a ChatGPT answer — because that is what the prospect actually reads.”— Head of Digital, leading Moroccan telco (anonymised)
The GEO playbook for Moroccan brands
- Audit entity disambiguation. Does the brand name resolve cleanly to your company on all four engines? Homonym collisions are the silent visibility killer.
- Publish structured, quantified, dated facts. Generative engines prefer numbers they can attribute. 'Founded in 1902, 14,000 employees, present in 26 countries' beats 'a leading Moroccan group'.
- Maintain the Wikipedia article with ten-plus independent secondary sources. This is the single largest lever for Claude and Gemini.
- Displace stale negative citations with fresher, higher-authority coverage. A 2026 press release does not erase a 2019 incident, but a 2026 independent investigation does.
- Run a quarterly prompt battery. AI-engine training surfaces shift; a static snapshot ages in one quarter.
Why this matters now
The share of B2B and consumer research journeys that touch a generative engine crossed 40% in our December 2025 panel. For graduate-talent decisions the share is higher. A brand that is invisible or mis-described in these answers is losing decisions it cannot see. The companies that treat AI visibility as a managed asset — like OCP, Maroc Telecom and Attijariwafa — are quietly compounding an advantage that is invisible in traditional search analytics.
Harch Atelier's AI Visibility Audit runs a 240-prompt battery across eight engines, scores citation rate, sentiment and factual accuracy, and delivers a GEO action plan ranked by expected impact. The audit is repeated quarterly to track engine drift.