Periscop
Periscop measures how ChatGPT, Claude, Gemini, Perplexity, Mistral and Google's AI overviews talk about a brand: Share of Model, mention rate, sentiment, hallucinations. Built for marketing teams, agencies and SMEs, hosted in Switzerland and billed in pay-as-you-go credits.
What is Periscop?
Periscop is a Swiss SaaS tool in the Generative Engine Optimization (GEO) category: it measures how visible a brand is inside generative AI assistants. It is published by Ergopix Sàrl, a web agency based in Vevey that also publishes Plumeo and Sezam. The tool queries six engines (ChatGPT from OpenAI, Claude from Anthropic, Gemini from Google, Perplexity, Mistral, and Google's AI overviews captured through DataForSEO) and reports five indicators. The central one is the Share of Model, a transposition of advertising share of voice: the numerator counts AI answers citing the brand, the denominator the total market mentions, brand plus competitors. The worked example given on the site is 18 citations out of 120 market mentions, a 15% Share of Model on ChatGPT. Four further measures sit around it: mention rate, sentiment (positive, neutral, negative or mixed), the sources the models cite, and hallucinations. Ergopix claims prompt quality as the product's signature and documents its method publicly. It runs in two passes. First, an AI reads the brand's website and produces a structured brief: personas, likely competitors, USP, geographic area, positioning, topics to avoid. A human then reviews and corrects that brief, about ten minutes, presented as the safeguard of the whole method. The second pass generates around twenty prompts from the validated brief under stated rules: at least three per persona, at least two carrying a differentiator, at least five geolocated, a mix of six registers and coverage of six search intents. The reasoning is that real users type short and in lowercase rather than asking polite questions, so measuring formal prompts amounts to monitoring a fictional reality. Hallucination detection is illustrated by a captured case in which Sonar Pro invented a restaurant, its address, its phone number and its domain, none of which existed. The home page states that four brands in ten face at least one factually wrong claim in AI answers. Measurements run on demand or on a weekly or monthly schedule, disabled by default on every new tracking setup. The output is a live dashboard, PDF reports by email and change alerts. There is no installation and no technical integration; access is invitation-only, with 100 free credits at sign-up and no bank card required. The interface is available in French, English, German and Italian.
What it does
- Measure a brand's Share of Model, model by model
- Query six engines: ChatGPT, Claude, Gemini, Perplexity, Mistral and Google's AI overviews
- Track the mention rate and its evolution, per prompt and over time
- Detect the competitors cited by the models, including undeclared ones
- Analyze the sentiment of the answers (positive, neutral, negative, mixed)
- Flag hallucinations, meaning wrong facts stated about the brand, and suggest corrective actions
- Surface the sources the models rely on, then send scheduled PDF reports and visibility alerts by email
When to use Periscop / When not to
A quick filter to help you decide if Periscop is the right fit.
When to use Periscop
- In-house marketing teams that need to know whether their brand is named by AI assistants, model by model
- Agencies and independent consultants reporting on generative-search visibility for client brands
- Swiss or multilingual brands addressing French, German, English and Italian-speaking markets
- Local businesses such as restaurants, hotels and shops exposed to invented addresses, opening hours or phone numbers in AI answers
- Fiduciaries, professional firms and industrial SMEs that want Swiss hosting and nLPD compliance without a monthly subscription
When not to use Periscop
- Teams looking for classic keyword SEO: Periscop does not replace traditional search-ranking tools
- Anyone wanting a free dashboard just to take a look, since every measurement really queries the models and consumes paid calls
- Brands that need real-time social media monitoring, which is the job of social listening tools
- Businesses with no website or online presence yet, because there is nothing for the models to cite
- Teams that expect self-service sign-up, a public API or a mobile app: access is invitation-only and Periscop offers neither API nor app
How to use Periscop
A typical end-to-end flow, from setup to results.
- Request an invitation code by email to hello@periscop.ch or through the contact form
- Create the account with the code, your company name, email and display name, then confirm the verification code sent by email: about two minutes, 100 free credits, no bank card
- Open the five-step configuration wizard covering identity, client brief, brand terms, prompts and competitors
- Let the AI analyze your website and propose a structured brief (personas, USP, geographic area, positioning)
- Review and correct the brief, adding your own knowledge of the market: five to ten minutes
- Validate the roughly twenty generated prompts, asking for variations where they read wrong
- Declare your main competitors
- Run a measurement on demand, or switch on weekly or monthly scheduling, which is off by default
- Read the results in the live dashboard and receive the PDF reports by email
- Top up your balance from the credits page, choosing a pack or any free amount in multiples of 10 CHF
Pros & Cons
Pros
- Per-model granularity: a score for each engine rather than a single blended figure, across six engines including Google's AI overviews
- The prompt generation method is documented publicly and includes a mandatory human review of the brief
- Hallucination detection comes with corrective actions, and the site shows a concrete captured case
- Cited sources are surfaced, which turns the measurement into something actionable rather than a bare observation
- No subscription and no commitment: credits are bought by the unit, never expire, and are automatically refunded when a scheduled action fails
- Swiss publisher and Swiss hosting, dual nLPD and GDPR regime, subprocessors named individually, self-hosted cookieless Matomo with no consent banner
- Setup in around ten minutes with no installation, a four-language interface (FR, EN, DE, IT), scheduling off by default and email alerts below 100 credits and at zero
Cons
- Access is invitation-only: nobody can sign up unaided
- No permanent free plan, and the 100 credits given at sign-up are a one-off allowance covering roughly five prompt measurements
- High real cost for dense tracking: 20 credits (20 CHF) per prompt measured on one model, 100 credits per monthly report, 250 credits per recommendation pack
- Credits cannot be refunded once purchased
- No public API and no mobile app, so no way to feed the data into another tool or an in-house pipeline
- No DPA, no EU representative under Article 27, no named DPO, no minimum age stated, and nothing said about whether customer data trains models or how to opt out
- Thin track record: domain registered in May 2026, no published incorporation date, no social presence, and a demo and case study built on an explicitly fictional brand (Café Lumen)
Pricing & Plans
Periscop does not offer a permanent free plan. Every new account receives a one-off allowance of 100 credits at sign-up, with no bank card required. Usage is then billed in credits, one credit being equal to one Swiss franc (CHF), with no subscription, no setup fee, no tracking fee and no mandatory package. The lowest paid entry point is the Starter pack at CHF 100; once registered, a customer may also top up any free amount in multiples of CHF 10. Swiss VAT at 8.1% is included in the displayed price for Swiss customers. Credits carry no expiry date but are not refundable once purchased.
- CHF 100 for 100 credits
- covering roughly 5 prompt measurements or 1 automated monthly report
- CHF 500 for 500 credits
- covering roughly 25 prompt measurements
- 5 monthly reports or 2 AI recommendation packs
- presented on the site as the most chosen and recommended option
- CHF 1
- 000 for 1
- 000 credits
- covering roughly 50 prompt measurements
- 10 monthly reports or 4 AI recommendation packs
- any top-up in multiples of CHF 10
- available from the credits page once registered
- 20 credits to measure one prompt on one model
- 250 for an AI recommendation pack
- 100 for an automated monthly report
- 30 for a client brief analysis
- 20 for a set of around 20 prompt suggestions
- 10 for variations of one prompt
Data, GDPR & hosting
A consolidated view of how Periscop handles your data.
GDPR overview
The privacy policy, last updated in June 2026, claims a dual regime: the Swiss Federal Act on Data Protection (nLPD) and the GDPR "where it is applicable". Concrete elements are present: a named controller with a postal address (Ergopix Sàrl, Avenue Nestlé 22, 1800 Vevey), enumerated legal bases (performance of the contract, consent for the contact form, legitimate interest), the full set of rights (access, rectification, erasure, objection, restriction, portability) exercised by email to hello@periscop.ch, subprocessors named one by one, transfers outside Switzerland framed by standard contractual clauses or equivalent mechanisms, and self-hosted cookieless Matomo analytics with no consent banner. Recourse is directed to the Swiss PFPDT and to the competent supervisory authority for EU residents. Two gaps are explicit: no EU representative is designated under Article 27 and no data protection officer is named. No data processing agreement is mentioned anywhere.
Who owns the data?
Under article 4 of the terms, Ergopix Sàrl keeps exclusive ownership of the Periscop platform, its code, interface, brand and visual identity; the customer receives a personal, non-exclusive right of use with no transfer of ownership. Data and content supplied by the customer remain the customer's property. Ergopix acts as data controller for account details (name, email, company), tracking setup (prompts, brand terms, website) and analysis results, and states that no data is sold or transferred for commercial purposes. Measurement queries are passed to third-party AI providers, namely OpenAI, Anthropic, Google, Mistral AI and Perplexity, alongside DataForSEO, Infomaniak and Stripe. The site never says whether customer data is used to train models.
Reuse rights
Customers keep ownership of the data and content they supply, and the terms place no license or reuse restriction on the reports Periscop produces. Reports are delivered as PDFs by email and can be downloaded from the customer area, while results stay consultable in the live dashboard, so a customer may read, share, quote or reuse them internally or with clients without asking Ergopix for permission. The documented data flow also runs inward: a file exported from Sezam can be imported into a Periscop tracking setup. The limits are about reliability rather than rights. Article 6 of the terms states that neither the completeness nor the accuracy of the analyses is guaranteed, because the results depend on third-party AI services over which Ergopix has no control, and that Ergopix accepts no liability for decisions taken on the basis of those analyses.
Data retention & training
Hosting summary
Jurisdiction is Swiss: the terms apply Swiss law with exclusive jurisdiction in Vevey, and the privacy policy operates under the nLPD, with the GDPR where applicable. Ergopix Sàrl, Avenue Nestlé 22, 1800 Vevey, is the data controller. The site and the application are hosted by Infomaniak Network SA, Rue Eugène-Marziano 25, 1227 Les Acacias (Geneva), Switzerland, and the policy states that the data is hosted in Switzerland. Independent network checks agree: the site resolves to 84.234.23.187, in Geneva, on AS29222 Infomaniak. Subprocessors are named individually: Infomaniak for hosting and email, the AI providers OpenAI, Anthropic, Google, Mistral AI and Perplexity, DataForSEO, and Stripe Payments Europe Ltd in Ireland for payments, acting as an independent controller. Transfers outside Switzerland, notably to the European Union and the United States, are framed by standard contractual clauses or equivalent mechanisms. Declared security measures: third-party API keys encrypted in the database with libsodium, strictly isolated per-customer access, HTTPS for all exchanges, and a PCI-DSS Level 1 certified payment provider. Audience measurement uses Matomo, self-hosted by Ergopix, with no cookie and no third-party sharing.
Things to keep in mind
Risks and trade-offs to weigh before adopting Periscop.
- Results come from non-deterministic third-party models, so two identical measurements can diverge, and the terms guarantee neither completeness nor accuracy
- The publisher admits the metric rests on assumed demand rather than observed demand; steering a strategy on it means accepting that gap
- Over-optimization risk: writing to please the models rather than the customers who eventually read the answers
- Costs drift fast at 20 CHF per prompt and per model, credits are not refundable once purchased, and the product is young and invitation-only, so its longevity is not established
- Prompts and reports describe a brand's positioning, competitors and customers, which the site itself flags as sensitive, and the queries travel to AI providers located outside Switzerland, in the European Union and the United States
- Nothing on the site says whether customer data is used to train models or how to object, and there is neither a DPA nor an Article 27 representative for a GDPR-bound buyer
- Delegated judgment: the human review of the brief is the safeguard the whole method rests on and is exactly the step a busy user will rush, while hallucination detection only recommends, leaving the correction on public sources entirely manual
Setup & Integrations
Technical difficulty
Very low. Periscop needs no installation and no technical integration: nothing to install, no API key to supply, no tag to place, no access to grant and no sales call. Account creation is announced at two minutes, followed by a guided five-step wizard with AI assistance at each stage; the site advertises three steps and ten minutes end to end. The only prerequisite is non-technical, obtaining an invitation code. The real effort is editorial rather than technical, since reviewing the AI-generated brief and validating the prompts takes five to ten minutes and assumes you know your market.
Deployment
Integrations
Supported languages
Behind Periscop
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Which AI engines does Periscop track?
What is the Share of Model?
Is there a free plan?
How do credits work?
Do credits expire?
Why do I need an invitation code?
How long does it take to get started?
Where is the data hosted?
Is there an API or a mobile app?
How is VAT handled?
Should you pick Periscop?
Periscop stakes out a clear position in a young category. GEO tools are appearing fast, and few of them name and define their central metric in public; Periscop does, with the Share of Model, its numerator, its denominator and a worked example. Its differentiators are real rather than cosmetic: per-model granularity instead of one blended score, a prompt generation method documented step by step with a mandatory human review, hallucination detection that arrives with corrective actions, and the domains the models cite surfaced alongside the numbers. The Swiss anchoring is used openly as a commercial argument, with Infomaniak hosting, a dual nLPD and GDPR regime and subprocessors named one by one, which matters to buyers for whom that is a contractual prerequisite. The business model is unusual for the category: credits bought by the unit, no subscription, no expiry, automatic refunds when a scheduled run fails. It suits an occasional user well and dense tracking poorly, since 20 CHF per prompt per model adds up quickly and credits are not refundable. Maturity is the reservation. The domain was registered in May 2026, sign-up is invitation-only, and both the demo and the case study rest on an explicitly fictional brand, Café Lumen, so those figures are illustrations, not customer results. For a demanding European buyer the compliance file has gaps: no DPA, no Article 27 representative, and nothing said about whether customer data trains models or how to object. Against that, the site shows an unusual willingness to argue against itself, with a page listing the cases where the tool is the wrong choice and an open admission that it measures assumed demand rather than observed demand. That candor is worth something when judging what the numbers actually mean.
- Choosing a selection results in a full page refresh.
- Opens in a new window.