Coaley Peak GEO Research Centre Engine guide
Seznam How Czechia's own search engine writes AI answers, and how to be the source.
Seznam is the Czech search engine that never gave way to Google. It now writes AI summaries with its own language model, SeLLMa, and runs an assistant, Seznam Asistent, that cites its sources. This guide covers its history, the four layers behind its answers, SeznamBot, where it crawls from and how to opt out.
Independent guide: Coaley Peak is not associated with Seznam.cz, a.s. Trademarks
- Owner
- Seznam.cz, a.s.
- Launched
- 1996 as a catalogue; own search engine March 2005; AI summaries January 2025
- Home market
- Czechia: 15% of searches in September 2026
- Languages
- Czech first
- Open weights
- No: SeLLMa is in-house and closed
- Crawlers
- SeznamBot, plus the Seznam-Extended AI opt-out token
Overview
What Seznam is, and who uses it
Seznam.cz began in 1996 as a Czech web catalogue, started by Ivo Lukačovič with CZK 50,000, and it is still privately owned by him. It added full-text search in 1997 and launched its own search engine in March 2005. Thirty years on, Seznam says its homepage, news, email, maps and business directory reach almost 90% of the Czech internet population.
In search itself Google leads, but Seznam keeps a larger share than any other national search engine in the EU: 15.05% of Czech searches across all platforms in September 2026, 16.59% on desktop and 13.75% on mobile, according to StatCounter. For a brand selling into Czechia, that is roughly one search in seven that Google's tools never show you.
Seznam has rebuilt its results around AI. Since January 2025 it has shown AI summaries written by SeLLMa, its own language model of more than 70 billion parameters, built mainly for Czech. In November 2025 it launched Seznam Asistent, a conversational assistant that answers from Search, news sites and Seznam's own services, and in May 2026 opened it to every logged-in user. Seznam's newer help pages call it Seznam AI.
That makes Seznam one of the few search engines where the whole chain, from crawler to index to model to answer, is built and run by one company outside the US. It also documents more of that chain than most: it publishes its crawler's IP ranges, a robots.txt token for opting out of AI answers, and how its summaries choose their sources.
Timeline
Seznam, from launch to today
Ivo Lukačovič launches Seznam.cz as a Czech web catalogue, with start-up capital of CZK 50,000.
Seznam adds Kompas, a full-text search indexing 1.2 million sites.
The Seznam Firmy business directory, forerunner of Firmy.cz, launches with 55,000 records.
Seznam launches its own search engine, with a database of 30 million Czech web documents.
Seznam presents SeLLMa, an in-house language model of more than 70 billion parameters built mainly for high-quality Czech.
Seznam explains its SeLLMa search summaries: about 4.5% of search traffic across all summary types, drawing on more than 70,000 domains so far.
AI summaries of news articles appear on the Seznam.cz homepage, written by Seznam's own AI.
AI summaries reach 6.2% of search traffic, and Seznam says publishers can opt out of being summarised.
The Seznam Asistent beta opens to selected users, with source citations, running on SeLLMa alongside OpenAI models hosted on Microsoft Azure in European data centres for the time being.
Seznam marks 30 years and says it reaches almost 90% of the Czech internet population.
Seznam Asistent opens to all logged-in users.
Seznam publishes a help page on keeping a site out of its AI answers with the Seznam-Extended, AI and MachineLearning robots.txt tokens. It was last updated on 22 September 2026.
Up to two Sklik ads, labelled 'Sponzorované výsledky' (sponsored results), start appearing under Seznam Asistent answers.
Things to know
Things to know about Seznam
Seznam's results-page AI summary is built from the top five organic results returned by its own Search API, and each point is attributed to its source. Ranking in Seznam's top five is the way in.
Seznam documents its own AI opt-out: a robots.txt group for Seznam-Extended, AI or MachineLearning with Disallow: / keeps a site out of its AI answers. The help page is in Czech only.
SeznamBot ignores X-Robots-Tag HTTP headers. Put noindex in an HTML meta robots tag and leave the page crawlable, or Seznam will never see the instruction.
Throttle SeznamBot with Request-rate, not Crawl-delay. The documented minimum differs between Seznam's English page (one document every 10 seconds) and its Czech page (one every 30 seconds).
Seznam supports IndexNow at search.seznam.cz/indexnow, with no cap on URLs. Seznam Webmaster's add-URL API is faster, indexing up to 500 URLs a day within minutes.
Seznam Webmaster (reporter.seznam.cz/wm) verifies a site with a seznam-wmt meta tag or file, and reports crawled, indexed, redirected and error URLs, with an API.
AI summaries mostly appear for informational, non-commercial queries that have no other direct answer: how-tos, home and garden, pets, travel, education and health.
Seznam Asistent draws on Firmy.cz, Zboží.cz, Počasí.cz, Sport.cz, public transport timetables, TV listings, news sites and Search. Firmy.cz lists more than 740,000 businesses, and a basic profile is free.
Since June 2026, Sklik search and smart-campaign ads can appear under Asistent answers automatically. Advertisers cannot target or exclude that placement separately.
Seznam ignores the meta keywords tag and judges relevance on the title and the text of the page.
The four layers
How Seznam builds an answer
Every AI answer is made in four layers: what the model learned in pre-training, how it was tuned, what it retrieves live, and how it puts the answer together. Here is each one for Seznam, and what it means for a brand that wants to be cited.
Layer 1
Pre-training: the base (foundation) model
SeLLMa is Seznam's own family of language models. Seznam described a model of more than 70 billion parameters, built mainly for high-quality Czech, in October 2024, and says SeLLMa now runs in more than 60 applications across its services. The weights are closed, and Seznam publishes no knowledge cut-off or training data. Czech press reports that SeLLMa was built by continuing the pre-training of open models, Meta's Llama 3.1 and earlier Mistral, on Czech text; Seznam's own documentation does not say so.
What it means for you: What SeLLMa knows about your brand comes from Czech-language text and, if the press reports are right, from what the open models beneath it learned first. A brand that is well described in Czech, on Czech sites, gives the model something to learn. One known only in English gives it very little.
Layer 2
Post-training: tuning and alignment
Seznam publishes nothing about how SeLLMa is tuned or aligned. Its help pages carry the usual caveats: AI output can be wrong, biased or incomplete, and is not medical, legal or financial advice. When the Asistent beta opened in November 2025, Seznam said it was also using OpenAI models run on Microsoft Azure in European data centres for the time being, with the aim of running only its own models in its own data centres. The May 2026 rollout mentions only SeLLMa.
What it means for you: Tuning decides tone and caution rather than which brand is named, but it does mean answers on regulated topics will hedge. Every AI element has a thumbs-down button that opens a feedback form: the only documented way to report an answer that gets your brand wrong.
Layer 3
Retrieval: live search
Seznam answers from its own index, built by SeznamBot. For a results-page summary it calls its internal Search API, takes content from the top five organic results, filters it, and has SeLLMa write sourced bullet points. Trending-topic summaries use Seznam's News API. Seznam Asistent reaches wider: Search, news sites, Firmy.cz, Zboží.cz, Počasí.cz, Sport.cz, public transport timetables and TV listings. A separate 'direct answer' element answers from the model alone, with no web sources.
What it means for you: Most Seznam GEO is won here. If a page is not crawled and indexed by SeznamBot, and does not rank in Seznam's top five for the query, it cannot be a source for the summary. Classic Seznam SEO, IndexNow and a complete Firmy.cz profile are the levers that move it.
Layer 4
The answer: what people read
Summaries are short bullet points, with sources shown beside each point or beneath the summary, typically three to five of them. Asistent answers cite their sources too. Logged-in users are invited to continue in the assistant from summaries, featured snippets and news teasers. Since June 2026 up to two Sklik ads can sit under Asistent answers, and Seznam's shopping offers have appeared there since May 2026.
What it means for you: Being cited is visible: your domain sits next to the point it supports. Write pages whose key facts can be lifted into a single Czech bullet point with your name attached, and track which of your pages Seznam actually cites, not only where they rank.
Crawlers
Seznam's crawlers and user agents
The user-agent tokens to look for in your server logs and to name in robots.txt.
| User-agent token | What it does | robots.txt |
|---|---|---|
| SeznamBot Documentation | Seznam's main search crawler. It finds new pages and refreshes known ones, and its index feeds organic results, AI summaries and Seznam Asistent. User agent: Mozilla/5.0 (compatible; SeznamBot/4.0; +https://o-seznam.cz/napoveda/vyhledavani/en/seznambot-crawler/). | Follows robots.txt, including Allow, Disallow, the * and $ wildcards, Request-rate and Sitemap. Ignores X-Robots-Tag headers. |
| Seznam-Extended Documentation | Not a crawler: a robots.txt token that keeps a site out of Seznam's AI answers. Seznam also honours the generic tokens AI and MachineLearning for the same purpose. | Any one of the three tokens with Disallow: / is enough, Seznam says. |
| SeznamBot (test version) Documentation | A test build of SeznamBot, with its own IP addresses and fulltextrobot-test-*.seznam.cz reverse DNS. | Not published separately |
| FreshBot Documentation | A fast-turnaround crawler for pages and RSS feeds where fresh content appears, mainly news sites. Its own robots.txt token and user agent are not published. | Not published |
| IndexNow key check Documentation | Fetches IndexNow key files to verify URL submissions to Seznam. No user-agent token is published for it. | Not published |
Seznam's Czech FAQ still shows an older user agent, SeznamBot/4.0-RC1 with a napoveda.seznam.cz link, so match allow-lists and log filters on the SeznamBot token rather than the full string.
Crawl sources
Where Seznam crawls from, and how to verify it
Seznam publishes SeznamBot's addresses as JSON at search.seznam.cz/ipranges/seznambot.json, last generated on 13 July 2026. Its help pages list production IPv4 addresses in 77.75.76.x, 77.75.77.x, 77.75.78.x and 77.75.79.x, and the IPv6 ranges 2a02:598:64:8a00::302:340/120 and 2a02:598:96:8a00::b00:80/118. RIPEstat shows the IPv4 block announced by AS43037, Seznam's own network; Seznam does not publish the ASN itself.
To verify a request, run a reverse DNS lookup. Every genuine SeznamBot IP resolves to a fulltextrobot-*.seznam.cz hostname, such as fulltextrobot-77-75-77-xxx.seznam.cz; look that hostname up again and check it returns the same IP. The test crawler uses fulltextrobot-test-*.seznam.cz hostnames, the addresses 77.75.73.26 and 77.75.72.26, and the IPv6 ranges 2a02:598:64:8a00::f00:0/123 and 2a02:598:96:8a00::1200:100/123.
IndexNow key checks come from 77.75.73.74 and 77.75.72.74. Anything calling itself SeznamBot from outside these ranges is not Seznam, and blocking it will not affect your Seznam visibility.
Opting out
How to opt out of Seznam, or limit what it uses
Opt out of AI answers in robots.txt
Add a group for Seznam-Extended, AI or MachineLearning with Disallow: /. Seznam says any one is enough, and that it will then not use the site as a source for AI answers. Its Czech help page does not say whether this also stops its models being trained on the site.
Official pageOr opt out with a meta tag, at a cost
<meta name="seznambot" content="nosnippet"> also stops AI answers, but it removes every text snippet, preview and rich result from your listings too. The robots.txt route is the gentler one.
Official pageTurn off featured snippets only
<meta name="seznambot" content="nofeaturedsnippet"> blocks featured snippets and leaves ordinary snippets alone. Seznam's own study found featured snippets raise average click-through by about 7%, so weigh it first.
Official pageBlock or slow SeznamBot
User-agent: SeznamBot with Disallow: / stops crawling altogether; Request-rate slows it instead, for example Request-rate: 300/1m. Seznam says changes usually take several days to be picked up, and several weeks on rarely visited sites.
Official pageRemove pages from the index
Return 404 or 410 (410 is removed faster), or add a noindex meta robots tag and leave crawling allowed. SeznamBot ignores X-Robots-Tag headers, and a robots.txt block would stop it reading the noindex.
Official pageRemove personal data from results
Ask the site owner to remove it first, then resubmit the URL through Seznam's add-page form. If that fails, email vyhledavani@firma.seznam.cz with the queries, the exact URLs and the reason for each.
Official pageCorrect your business details
Claim and edit your free Firmy.cz profile, or email info@firmy.cz. Profiles feed Seznam Search, Mapy.com, Seznam's other services and Asistent, so a wrong address or phone number spreads quickly.
Official page
Opting out of AI answers stops your pages being quoted, but it does not take them out of Seznam's index or rankings, and it cannot remove anything SeLLMa has already learned. Seznam's help pages also disagree on whether meta name="seznambot" works for index directives, so test any meta-tag rule before relying on it.
How we help
How we get brands cited in Seznam
The Coaley Peak GEO Research Centre researches and implements GEO for SEO agencies and in-house teams, with foreign markets as our speciality.
Seznam visibility audit
We check which of your pages SeznamBot has crawled and indexed, where they rank in Seznam's top five for the queries that trigger AI summaries, and which domains Seznam cites in your place.
Czech content written to be cited
With native Czech writers, we rework your key pages so their facts survive being cut down to one bullet point: clear claims, named products, dates and figures, in the Czech your customers search in.
Technical set-up for Seznam
IndexNow and Seznam Webmaster, the right robots.txt rules and meta tags (not X-Robots-Tag), and a crawl SeznamBot can finish, checked against Seznam's published IP ranges.
Firmy.cz and Seznam's own services
Asistent answers from Firmy.cz, Zboží.cz and Seznam's other services. We get your profiles complete and consistent, so the assistant has accurate facts about you to repeat.
Measurement across Seznam and Google
We track your citations in Seznam's summaries and Asistent alongside Google's AI Overviews in Czechia, so you see the share of Czech AI answers you win, not just your rankings.
More on our research, services and team at the Coaley Peak GEO Research Centre.
FAQ
Seznam GEO: questions
Does Seznam still matter in Czechia?
Yes. StatCounter put Seznam at 15.05% of Czech searches across all platforms in September 2026, and Seznam says its wider services reach almost 90% of the Czech internet population. For a Czech campaign that is too much demand to leave to tools that only see Google.
How does Seznam choose the sources for its AI summaries?
For results-page summaries, Seznam takes content from the top five organic results returned by its own Search API, filters it, and has its SeLLMa model write bullet points with each source attributed. To be cited, you need to rank in Seznam's top five for the query.
How do I stop Seznam using my site in AI answers?
Add a robots.txt group for Seznam-Extended, AI or MachineLearning with Disallow: /. A seznambot nosnippet meta tag also works, but it removes all your snippets and rich results too. Neither takes your pages out of Seznam's normal results.
What is SeLLMa?
SeLLMa is Seznam's in-house large language model family, presented in October 2024 as a model of more than 70 billion parameters built mainly for Czech. It writes Seznam's search summaries and runs in more than 60 Seznam applications. Its weights are not public.
How do I check that a visitor really is SeznamBot?
Check its IP against Seznam's published list at search.seznam.cz/ipranges/seznambot.json, then confirm that a reverse DNS lookup returns a fulltextrobot-*.seznam.cz hostname which resolves back to the same IP.
Does Seznam support IndexNow?
Yes, at search.seznam.cz/indexnow, with no cap on the number of URLs; Seznam says selected URLs typically reach results within a week. Seznam Webmaster's add-URL API is faster, indexing up to 500 URLs a day within minutes.
Are there ads in Seznam's AI answers?
Since June 2026, up to two Sklik ads labelled 'Sponzorované výsledky' (sponsored results) can appear under Seznam Asistent answers. Advertisers cannot target or exclude that placement separately; its results are reported within campaign statistics.
Want a free chat about Seznam?
Tell us which market and brand you are working on. If we can help, we will set up a free, no-obligation chat with Stephen or Ethan.
More engine guides
Related engines
Independence and trademarks. Seznam, Seznam.cz, SeznamBot, Firmy.cz, Sklik and Mapy.com are trademarks of Seznam.cz, a.s. or its affiliates. Coaley Peak is independent: we are not associated with, endorsed by or a partner of Seznam.cz, a.s. This guide is our own research from public documentation and reputable reporting, checked in October 2026.
Reviewed by Stephen Grindley, research centre lead. Last checked .
Document reference: ISO_webpage_geo_engine_seznam_v1
Last modified: 7 October 2026
GEO Research Centre·Engine guides·Seznam