Generative engines have crossed the line from novelty to infrastructure. When a prospect, a journalist, an investor or a graduate asks 'is [company] reliable', the first answer they read is no longer a search result — it is a generated paragraph. That paragraph is the new front page, and it is sticky: once an answer enters an engine's citation surface, it recirculates for weeks. The Harch AI Reputation Index measures who is winning this new channel, and why it matters more each quarter.
The query-volume shift
Our panel data shows generative-engine query volume for Moroccan company names up 4.1× year-on-year in Q2 2026. The absolute number now approaches 35% of traditional search volume for the same queries, and for evaluative prompts ('is X reliable', 'is X safe', 'should I bank with X') the generative share is over 50%. The trajectory, not the level, is the strategic fact.
Traditional search is not collapsing — it is plateauing while generative grows through it. The implication is additive, not substitutive: a company must manage both surfaces. But the generative surface is where the marginal decision is now made, and where the marginal reputation dollar produces the highest return.
Why citations are sticky
A search result can be displaced in days by a fresher, better-linked page. A generative-engine citation is stickier because it enters the engine's training and retrieval surface, where it is reinforced by repeated retrieval. Our tracking shows that once a negative citation enters ChatGPT's answers, it persists for an average of 14 weeks; on Perplexity, 9 weeks; on Gemini, 11 weeks. Displacement requires a denser, fresher, higher-authority counter-narrative than the original citation — a higher bar than traditional SEO ever set.
Citation rate benchmarks
The Harch AI Reputation Index tracks citation rate — the percentage of relevant prompts in which an engine cites the brand — across eight engines and 100 Moroccan companies. The chart below shows the citation-rate benchmark for the top quartile, median and bottom quartile, by engine.
The spread between the top quartile and the median is the strategic fact. The leaders are not slightly more cited — they are roughly 50% more cited. That gap reflects the three drivers we identified in the AI visibility audit: structured-data richness, recency and volume of indexed press, and Wikipedia presence with independent sources.
What shapes the answer
A generative engine's answer to 'is [company] reliable' is shaped by three layers. The retrieval layer pulls the freshest, most authoritative sources it can find. The synthesis layer composes them into a paragraph, weighting recency, authority and consistency. The training layer encodes the composite into the model's parametric memory, where it persists even without fresh retrieval. A company that is invisible at the retrieval layer is invisible in the answer; a company that is visible but inconsistent produces a mixed answer; a company that is visible, consistent and fresh produces the answer it intends.
The AI reputation moves
- Measure the baseline. Run a 240-prompt battery across eight engines and code citation, sentiment and accuracy. You cannot move what you have not measured.
- Close the Wikipedia gap. A well-maintained article with ten-plus independent sources is the single largest lever for Claude and Gemini.
- Publish structured, dated, quantified facts. Generative engines prefer numbers they can attribute.
- Displace stale negatives with fresher, higher-authority coverage. A 2026 independent investigation displaces a 2019 incident; a 2026 press release does not.
- Re-run quarterly. The training surface shifts; a static snapshot ages in one quarter.
“We spent ten years on SEO and three months on GEO. The three months moved our AI-engine answer more than the ten years moved our search ranking.”— Head of Digital, Moroccan retailer (anonymised)
Why this is the year
The 4.1× growth rate means the generative channel is compounding. Every quarter a company delays measurement and action, the gap to the leaders widens — and the stickiness of the existing answers means the gap is harder to close later. The companies that move in 2026 will set the citation baseline the engines train on for the next cycle. The companies that wait will spend 2027 displacing answers they could have shaped.
Harch Atelier's AI Reputation Index runs the 240-prompt battery across eight engines, scores citation rate, sentiment and accuracy, and delivers a quarterly GEO action plan. The index is the foundation of our AI Visibility Audit.