State of the
Moroccan Web.
Connections, technology, hosting, language
and content across Moroccan websites.
A September 2026 measurement of how the Moroccan web is structured, what it depends on, and what its pages contain. The Omneity Moroccan Web Observatory produces this report from Omneity Maktaba, the first Moroccan-focused search index.
searchable pages
in the study’s wider corpus
Moroccan source websites
in the link analysis
recorded hyperlinks
from Moroccan source websites
The wider corpus includes all sources; the Moroccan link analysis covers 128,335 pages. This is a study sample, not a census of the Moroccan web. About the study
Start exploring / 500 destinations
Explore the top 500 destinations
Where do Moroccan websites point? Explore the 500 highest-scoring destinations. Filled gold circles mark known Moroccan websites. Outlined circles mark other destinations. Lines connect Moroccan sites with observed hyperlinks between them. Deeper amber marks sites with more incoming and outgoing connections.
The destination cloud needs a browser with JavaScript and canvas support. Read the public ranking.
Select a Moroccan site to trace its links · Drag to explore
Connections of marchespublics.gov.ma
4 total · incoming + outgoing
One connection means at least one recorded hyperlink, not an endorsement or a measure of traffic. Only pairs within this top-500 selection are shown.
Three views of the Moroccan web
Connections and destinations
See how websites link internally, across the Moroccan web and beyond it.
InfrastructureTechnology and hosting
Measure common services, networks, endpoint countries, IPv6 and TLS.
ContentLanguage, terms and topics
Compare declared and classified language, then explore common phrases and tags.
01 / Connections
Internal and cross-site links
Most links help people move around the website they are already on. Among links that leave a website, destinations outside the known Moroccan set dominate.
Where all links lead
When a link leaves the website
point outside the known Moroccan set.
13.65% point to another known Moroccan
website.
Homepages have a higher local link share
On homepages, the local share of cross-site links rises to 29.2%. Repeated social links on deeper pages may help explain the difference. A homepage link and a repeated footer link do not carry the same meaning.
link to another known Moroccan site
4,954 of 17,571 source websites have an observed link to another known Moroccan site. Most have no observed local cross-site connection.
of local connections are reciprocated
The graph contains 12,778 distinct directed connections between Moroccan domains; 1,336 have a reverse link. Missing pages can hide real connections, so this does not establish social or institutional isolation.
02 / Destinations
Most-linked global platforms
Social, video and messaging platforms lead the destination ranking. More than half of the observed Moroccan source websites link to Facebook.
Leading linked destinations
Types of links
A social-profile link offers a place to follow or contact an organisation. A payment service, login provider or embedded application can play a more essential role. The presence of a link alone does not show how dependent a website is on that destination.
Referral coverage is not traffic
These figures show how widely destinations are referenced, not how many visitors they receive. Share buttons, footers and provider credits can all contribute.
Domain and company boundaries
Facebook, Instagram, WhatsApp and wa.me are separate destinations. Adding their percentages would count some source websites more than once.
Explore the top 1,000 destinations →03 / The local web
Most-linked Moroccan websites
Public-service portals appear alongside development, hosting and search-optimisation providers. Links can reflect navigation, attribution, suppliers, sources or editorial references.
Leading known Moroccan destinations
Why service portals appear
Government and administrative domains receive links from related institutions, documents, forms and public-information pages. The ranking measures how widely those links appear across source websites.
Why technology providers appear
Development, hosting and search providers often receive site-wide footer credits. One supplier relationship can therefore create links across many pages, while the score still counts each source website once.
The study identifies domain relationships. It does not classify every link by purpose or verify the ownership of every destination.
04 / Technology & resources
Third-party services and web technology
Fonts, libraries, maps, tags and commerce tools connect websites to services beyond their own domains. These relationships can affect appearance, functionality, privacy and resilience.
Common resource declarations
Server response labels
Publishing & commerce
Shopify asset declarations appear on 2.9% of the resource sample; WordPress-related s.w.org references on 4.1%. These are selected service signals, not estimates of merchant or CMS adoption.
Maintenance
Shared libraries can reduce duplicated work. Their versions, update paths and failure behaviour still matter. A resource hostname alone cannot establish software security.
Independent operation
A useful test is whether essential text, navigation and forms remain usable when an optional third-party service is unavailable.
05 / Privacy & tracking
Tracking and consent signals
Analytics-, advertising- and SDK-related hosts appear in the page code. That is evidence of a declared relationship—not proof that a tracker ran or that personal data was transferred.
Selected measurement and advertising signals
A tag manager is not a tracker count.
It can load different tags under different conditions. Static page code can miss later requests, first-party collection and server-side processing.
A consent tool is not proof of consent.
Selected consent-service declarations include CookieYes on 0.29%, cookielaw.org on 0.28% and Cookiebot on 0.14% of the resource sample. Custom and dynamically loaded tools are not captured by these selected hosts.
These percentages cannot estimate compliance or the share of websites that track visitors.
Declared service
Actual request
Information sent
Purpose & timing
What this measurement shows
It shows which selected service hosts appear in stored page resources. It does not observe a visitor session, inspect the information sent in requests or determine whether an individual website complies with privacy rules.
06 / Infrastructure
Hosting network concentration
Public network endpoints are concentrated. Almost two-thirds of the sampled domains touch at least one of the five leading networks.
Leading public endpoint networks
Five-network reach
61.9%Cloudflare · Hostinger · Hetzner · SoftLayer · OVH
12,836 of 20,751 sampled domains, with each domain counted once. This measures the public edge—not hosting revenue, every reseller, the hidden origin server, or database storage.
Moroccan network operators also appear in the sample. Network ownership and the country assigned to an IP address are different facts.
Resilience risks from shared providers
Common providers can make services easier to operate. They may also create shared points of failure. To assess the practical risk, test which user journeys depend on each supplier and whether alternatives exist.
07 / Endpoint geography
Hosting country estimates
A country attached to a public IP is one piece of evidence. It does not locate all of a website’s application servers, databases, backups, operators or other data locations.
Country labels of public endpoints
9.32% is an endpoint label.
Not a data-residency estimate.
Content delivery networks, shared addresses, redirects and lookup errors can separate the visible endpoint from the origin. Canada’s large share especially needs CDN-aware interpretation.
These figures do not establish that the remaining share of Moroccan websites is physically hosted abroad.
How to read the result
Treat the country label as the visible network location. The application, stored data, backups and operators can be elsewhere. Those layers require separate measurements.
DB-IP notes reduced coverage and accuracy in its Lite database.
08 / Security & readiness
IPv6 and TLS coverage
DNS and root responses reveal parts of the technical surface. A complete security or reliability assessment needs much more than these first observations.
advertise IPv6
6,962 of 20,751 sampled domains have an IPv6 address in the DNS answer. Advertisement does not prove working connections or equal performance.
have TLS verification marked true
18,864 of 20,751 successful root observations. Other values need protocol and error review; they do not automatically mean an invalid certificate.
Limits of these security signals
Certificate validity, DNS security, browser protections, software maintenance and application behaviour each answer different questions.
This edition does not grade websites for security or report exploitable vulnerabilities.
09 / Language & representation
Declared and classified language
A page’s language label and the language of its text can differ. We compare the declared label with Gherbal v4, Omneity’s language classifier, on the same text samples.
One root page and up to two inner pages per website. A limited sample of extracted text, not a count of every language offered by each site.
Declared language vs Gherbal v4
Same pages, same denominator. Each bar is a share of all 28,155 usable samples.
French
English
Arabic family
Spanish
Amazigh
Missing / unresolved
Do the labels match the text?
83.8%
agreement across 23,887 samples with a recognised declaration and a retained, single-language prediction. Arabic varieties count as one family here.
54.3% of samples marked English are classified as French (3,112 of 5,734). A template’s default language can differ from the published text.
Darija and Arabic varieties
Gherbal’s labels among retained Arabic-family samples. These are model predictions, not a human-validated dialect census.
Mixed writing systems
832 websites have a sampled text with at least 10% Arabic letters and 10% Latin letters. This can reflect multilingual text, brand names or navigation; it does not by itself identify Darija.
Classification checks and exclusions
The sample starts with 48,688 pages from 20,803 Moroccan websites. We use up to the first 6,000 characters of extracted main text per page. We exclude 12,156 samples with code or markup markers and 1,784 with common error, access-challenge or holding-page phrases. We then remove 5,334 samples with fewer than 40 words or 200 letters, plus 1,259 exact normalised text duplicates within the same website.
Gherbal v4 runs on each remaining text. Predictions below a model score of 0.80 are uncertain; this score is not a calibrated probability. A prediction also fails the script check when fewer than 20% of letters match its expected writing system. Other language predictions remain unreviewed, not automatically false.
- Low confidence: 1,838 samples
- Script conflict: 16 samples
- Other / unreviewed: 303 samples
- Mixed language: 398 samples
- Reviewed classification error: 3 samples
Mixed-language candidates have at least two recognised language families across 150-word windows, with the second family present in at least 20% of the windows. These stay outside single-language and agreement totals. All unresolved samples remain in the overall comparison denominator. Manual review also excludes three dialect predictions based mainly on lists of personal names. Short snippets were reviewed as a quality check, not as a representative accuracy evaluation.
These conservative filters can also exclude valid pages that quote code or error messages. They favour readily extracted text and do not remove every template fragment. The sample limits large publishers to three pages, but gives no guarantee of national coverage or equal language coverage. A missing language in this sample is not evidence of its absence from the Moroccan web.
10 / Words & topics
Common terms, tags and topics
Language tells us how pages communicate. Words, phrases and publisher labels help show what they discuss. These measures describe the sampled web, not the interests of all Moroccans.
Common words and phrases
What appears across websites, not what a few large sites repeat most often.
Share of 10,614 websites with a retained French text sample. Each website counts once per term.
- savoir faire10.9%
- casablanca maroc8.8%
- haute qualité8.3%
- temps réel7.0%
- besoins spécifiques6.9%
- service client5.3%
- gamme complète4.7%
- clients satisfaits4.5%
- solutions adaptées3.9%
- devis gratuit3.7%
- accompagnement personnalisé3.7%
- intelligence artificielle3.6%
Adjacent words only. Common stop words and selected navigation, cookie and social-media terms are filtered. Case and Arabic letter variants are normalised; words are not stemmed. Terms need at least 20 websites. These are frequencies, not trends or measures of public opinion.
Topic signals in the text
Share of 13,606 websites with at least two distinct topic cue words in a sampled page. Topics overlap.
Publisher-supplied tags
15.5% of sampled websites supply keywords; 2.6% supply a section label on at least one sampled page.
These labels describe how publishers label content. They can also contain search-engine optimisation terms. They are not independently verified topics.
Common keyword labels
Common section labels
Percentages use all 20,803 sampled websites, including those without tags. Each website counts once per label.
11 / Public dataset
The top 1,000
linked destinations.
Destinations linked to from the observed Moroccan web, ranked by referral coverage. Both Moroccan and international destinations can appear.
What the score means
A score of 10 means the destination is linked to by approximately 10% of observed Moroccan source websites. Each source website counts once per destination.
What it does not mean
It is not a traffic share, quality rating, recommendation or market share. Scores are rounded to two decimal places. Websites can refer to several destinations.
| Destination | Referral coverage score |
|---|---|
| facebook.com | 54.90 |
| instagram.com | 43.18 |
| linkedin.com | 31.29 |
| youtube.com | 23.56 |
| twitter.com | 17.75 |
| wa.me | 16.26 |
| google.com | 16.25 |
| whatsapp.com | 8.79 |
| tiktok.com | 7.30 |
| goo.gl | 6.72 |
| pinterest.com | 6.72 |
| cloudflare.com | 6.30 |
| x.com | 4.77 |
| wordpress.org | 3.23 |
| apple.com | 2.67 |
| telegram.me | 2.15 |
| shopify.com | 1.44 |
| youtu.be | 1.20 |
| wikipedia.org | 1.19 |
| heberjahiz.com | 1.14 |
| reddit.com | 1.02 |
| t.me | 1.02 |
| tumblr.com | 0.87 |
| hj.ma | 0.82 |
| intilaka.ma | 0.82 |
September 2026 edition. The download contains two fields only: domain and score. Search does not alter the download. Ranking is set before rounding;
equal underlying counts are ordered alphabetically.
12 / About the study
Methodology and limitations
Omneity Labs studies publicly accessible web pages through Maktaba. This edition reflects observations collected in September 2026, with a page cutoff of 9 September and root DNS observations from 6 September.
Which websites are Moroccan?
This edition includes .ma domains and a reviewed selection of Moroccan domains with other endings. The selection outside .ma is incomplete. Hosting country does not determine whether a website is Moroccan.
Why do the percentages use different samples?
Links use 17,571 websites with extracted hyperlinks. Resource charts use 18,572 websites with absolute resource references. DNS and server charts use 20,751 successful original root observations. Language and content use a separate, domain-capped text sample from the same page snapshot. Its sample sizes and exclusions appear in the language section. These groups overlap but are not identical.
What is a link?
A stored hyperlink, not a visit. Repeated links across pages remain in record-level shares. Destination scores instead count each referring website once, reducing the influence of large sites and repeated footers.
How is the public ranking prepared?
Destinations are grouped by domain. The score is 100 × referring source websites ÷ all observed Moroccan link-source websites. Only the top 1,000 eligible destinations are released, with scores rounded to two decimals.
Malformed names and public-suffix-only groups are excluded. Ties in underlying counts are ordered alphabetically. Domain grouping outside .ma can miss detail beneath multi-part suffixes; excluding those aggregate suffix entries does not recover the underlying websites.
What are the main limits?
The sample is not a census. Unseen or unextracted pages can hide links and services. Static declarations do not prove browser activity. IP labels do not prove physical hosting. A missing signal is not proof that a technology or practice is absent.
What does the network show?
The network shows observed directed links only when both websites are known Moroccan destinations in the top 500. Repeated hyperlinks collapse into one domain pair. These pairs are visible in the page; the top-1,000 download still contains only domain and score. Circle positions and line lengths do not measure geography or closeness. Colour shows incoming plus outgoing connections within this selected network, using a square-root scale. Two opposite-direction links count twice. Colours keep the same scale across filters; other destinations have no connection-based colour.
Data sources
- Web observations: Omneity Labs / Maktaba, September 2026. Public release: the destination ranking above.
- IP country labels: IP Geolocation by DB-IP, Country Lite, September 2026, CC BY 4.0. The Lite edition has reduced coverage and accuracy.
- Network labels: IPtoASN, September 2026, PDDL v1.0.
- Domain eligibility: Public Suffix List, used to exclude suffix-only entries from the public ranking.
An observatory by Omneity Labs
Tracking the state of the Moroccan web
The Omneity Moroccan Web Observatory measures how the web changes across editions. Maktaba provides the underlying index; this report publishes selected aggregate findings and the top 1,000 linked destinations.