Frequently asked
questions.

No. SEO works the list of results. Periscopy measures the generated answer, which is a different layer, with different sources and different winners. They coexist.

Coverage and reading. Most tools measure three or four engines, API only. We measure up to eleven engines every week, and on paid plans up to seven of them, depending on the plan, are read in the app itself, in a live session, and we read the data with you.

The first measurement is ready within hours. A trend needs four weeks of comparable data. Changing what the engines answer is months of work, and that is what the retainer and premium tiers are for.

Yes. Export everything to CSV whenever you want. The methodology and the platform are ours.

English by default. Other markets and languages via add-on or on the Business and Enterprise plans.

Yes. Every plan is monthly, with no minimum commitment and 30 days' notice.

Five questions from your category across four assistants, plus a 30-minute call where we read the result with you. No commitment, no card.

GEO (Generative Engine Optimization) is the work of making generative AI engines, such as ChatGPT, Gemini or Perplexity, know your brand, describe it correctly and recommend it when someone asks a question you are the right answer to.

It follows SEO in a world where a growing share of searches ends in a generated answer, with no click. SEO optimises to appear in a list of links; GEO optimises to be cited inside the answer itself. The techniques overlap in part (good content, structured data, authority), but the target and the measurement are different.

In practice, yes. GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) describe the same work: making a brand quotable by engines that generate answers instead of returning lists of links. The difference is mostly one of market and of period, since some countries and tools adopted one term and some the other.

This site uses GEO as the umbrella term because it describes the scope better. The work is not only for answer boxes but for any generative engine, from ChatGPT to Google AI Mode. If you see AEO, LLMO or AISO in an article or a proposal, look at what sits under the acronym: real measurement, technical and authority work, and honesty about what can be influenced.

No, and the confusion is common enough to show up in the answers of the engines themselves. Geomarketing is geographic market analysis: where to open a shop, how to segment campaigns by region. GEO, in the sense used here, is Generative Engine Optimization, the work of making a brand quotable by generative AI engines such as ChatGPT, Gemini or Perplexity.

The distinction matters because the two are searched for and sold under the same acronym. One question settles it: what exactly will be measured, and in which engines? If the answer does not include citations in AI engines, it is another discipline.

AI engines recommend companies they can read, understand and verify. In practice that takes four things: content on the site that answers the questions buyers actually ask, structured data (schema.org) that identifies the company without ambiguity, presence in the sources the engines cite when they search, and a consistent brand description across the platforms where models learn.

There are two distinct routes and both count. In training memory, the model answers from what it learned about the brand up to its cut-off date, and the entity matters most (the knowledge graph, consistent descriptions). In live web search, the engine searches and cites sources on the spot, and what matters is being on the pages it finds and treats as reliable. A company can be strong in one route and invisible in the other, and only a measurement that separates them shows which one is missing.

There is no way to pay for a place and no trick that replaces the work. What exists is method: measure where the brand stands today in each engine, find the questions with no dominant answer yet, and build a quotable presence before a competitor does.

Answering from memory, the model draws on what it absorbed in training: mentions on pages with authority, consistent descriptions of the company across several sources, presence in lists and comparisons. Searching the web, the process changes. The engine runs a search, reads a handful of pages and writes an answer citing the ones it found useful.

In both cases selection favours brands with a clear identity (who they are, what they do, for whom), verifiable proof (cases with numbers, open documentation) and presence in the right sources for the category. What holds a brand back: content that only exists after JavaScript runs, contradictory descriptions across platforms, and marketing claims with nothing underneath for a model to quote.

Answers also vary between engines and between weeks. The same question can be answered differently by ChatGPT, Gemini and Perplexity, which is why measurement runs continuously across several engines instead of drawing conclusions from a single search.

With three main metrics, measured over a fixed set of questions relevant to your brand. Citation rate is the percentage of answers that mention your brand. Share of voice is the slice of the category's mentions that is yours, against the other brands. Average position is the place you take when you appear in a list.

The word that matters is fixed: the same questions, in the same engines, week after week. A one-off search in a single engine on a single day is an anecdote, not a measurement, because engines vary between themselves and between runs. Only a time series with explicit denominators lets you say honestly whether visibility is rising.

In five steps. First, define a fixed set of questions, the ones your buyers ask, from generic to decision stage. Second, choose several engines and separate the two routes, training memory and live web search. Third, ask each question in each engine more than once, and record whether you are mentioned, recommended, and in what position. Fourth, calculate the metrics with an explicit denominator: citation rate, share of voice and position. Fifth, diagnose the cause of an absence, whether it lies in the entity, the content or the technical base.

A first snapshot can be done by hand and is worth doing to know where you stand. What hands cannot sustain is continuity: AI visibility changes week to week, and only a time series shows whether you are improving. Avoid the mistakes that invalidate an audit, which are a single engine, a single run, confusing a mention with a recommendation, and figures without a denominator.

It works in the sense that it is measurable. You define a fixed set of buyer questions, measure the percentage in which the brand appears, apply the fixes and measure again. If citation rises, it is working. If it does not, the data shows where the blockage is.

It is worth it for companies whose buyers ask AI assistants before deciding, which today includes B2B buyers, technical decision makers and consumers in research-heavy categories. Google AI Mode reached one billion monthly users in May 2026, so the question is no longer whether buyers use these tools, but whether your brand exists in the answers.

The answer can also be that it does not pay yet. If your category has residual demand in AI engines, or if the SEO base is not in place, a serious diagnostic says so. Investing in GEO without a foundation is paying to build the second floor first.

It depends on the starting point, but the honest pattern is this: technical fixes (schema, llms.txt, extractable content) show up in live-search answers within weeks, because the engines re-read the site on every search. Training memory moves more slowly, at the pace of each vendor's training cycles, and can take months to reflect new presence.

A sensible plan therefore runs in four stages: weeks 1 and 2 for the technical foundation, weeks 3 to 6 for content and sources, weeks 7 to 12 for external authority, and from 90 days on for entity consolidation and positioning. Each stage has its own actions and measurable signals, and weekly measurement shows progress without waiting for the end.

Distrust promises of results in days. The engines have no priority queue for sale, and anyone promising guaranteed positions in a non-deterministic system is promising what does not depend on them.

Yes, and more than that: it is the foundation GEO stands on. AI engines with live search read the same web Google indexes, and a site that is technically fragile, slow, or hides its content behind JavaScript fails in both worlds at once.

What changed is the ceiling. Classical SEO alone no longer guarantees presence where a growing share of decisions begins, because the generated answer replaces the list of links. The right reading is not to swap SEO for GEO but to build the AI layer on top of a well-made SEO base, as one strategy for being found.

It depends on who your buyer is, which is why measurement covers several. Google AI Mode carries the largest volume, because it sits inside the search everyone already uses. ChatGPT leads direct assistant use. Perplexity weighs with technical users and early adopters who search with citations. Claude, Gemini, Microsoft Copilot, Meta AI, Mistral and DeepSeek each carry their own audience.

The common mistake is to optimise and measure in a single engine. Answers vary considerably between engines, because they train on different data and search in different ways, and being strong in one and invisible in another is the pattern, not the exception. Periscopy measures up to eleven engines in parallel because your real visibility is the sum, not the best case.

It is one piece of the technical foundation, cheap to implement and with no downside. llms.txt is a markdown file at the root of the site that gives AI engines a curated index of the content: who you are, what you do, where the pages that matter live. The llms-full.txt variant delivers the main text of the site in a single read.

On its own it works no miracles, because it is an emerging standard and not every engine reads it today. But it costs an afternoon, and it reduces ambiguity for the engines that do read it, with the trend pointing up. The same technical hygiene includes a robots.txt that allows the relevant AI crawlers, and schema.org on the main pages.

For most brands, no. Blocking GPTBot, ClaudeBot or PerplexityBot removes the site from the sources the engines read, which means giving up on appearing in the answers where your customers are looking. It is a real trade between protecting content and being visible, and for anyone who lives on being found the arithmetic rarely favours blocking.

Fine-grained choice exists and should be used. The robots.txt file lets you decide crawler by crawler, and separate training bots from search bots. A publisher with paid content has reasons to block training; a services company that wants to be recommended has reasons to allow everything. The decision should be deliberate and revisited, not the default of a plugin.

First, the honest framing: the drop in clicks caused by generated answers is structural, not a technical fault on your site that an adjustment will fix. Google increasingly answers on the results page itself, and part of the traffic that used to exist does not come back. The strategy is not to recover the lost clicks but to exist inside the answers that replaced them.

In practice, there are three fronts. Be among the sources cited, because AI Overviews cite a handful of pages per answer, and extractable content with original data and correct schema competes for those places. Measure what changed, by separating in Search Console the losses caused by position from the losses caused by behaviour, and by following the impressions where the citation appears. And widen the engines, because whoever asks ChatGPT or Perplexity never goes through Google, and visibility in those engines is won separately.

Blocking AI crawlers in retaliation is the wrong answer, because it removes you from the answers without giving back the clicks. Presence in the answer became the asset, and the click follows when the intent is strong.

First, find the origin. If the error appears in answers from memory (no web search), the model learned outdated or contradictory information, and the fix is consistency: the same company description on the site, on LinkedIn, in directories and in every public source, reinforced by structured data. If the error appears in answers with live search, a specific page is feeding it, and that page is what has to be found and corrected.

Second, do not expect an instant correction in the models' memory, which moves at the pace of each vendor's training cycles. Live search corrects faster, because the engines re-read the web on every answer. In both cases the prerequisite is monitoring: without continuous measurement, a wrong description can circulate for months before anyone notices.

Yes, and it happens: prices that never existed, services you do not provide, headquarters in another city. Models generate plausible text, and when they hold little reliable information about a brand they fill the gaps with assumptions. Brands with little structured presence are the most exposed, precisely because they give the model less verifiable material.

The defence is twofold. Reduce the vacuum, because the more clear, consistent and structured information exists about the company (an extractable site, schema, identical descriptions on every platform), the less room is left for invention. And watch, by measuring regularly what each engine says, to catch inventions when they appear and correct the sources that sustain them.

Yes, on two conditions: real human review and substance of your own. Engines do not penalise text for having been written with AI help; they penalise thin, repeated content with no new information, whoever the author. A hundred pages generated in series on the same theme are spam to any modern engine, and Google treats mass production without value as abuse.

What makes content quotable is what only your company can add: original data, real experience, cases with numbers, positions taken. AI is good at giving shape, but the raw material the engines cite has to come from the brand. Used that way it speeds things up without risk; used as a volume factory, it builds a site the engines learn to ignore.

No, but what Wikipedia represents does matter: a verifiable entity, described by third parties, with sources. Models train intensively on Wikipedia and Wikidata, and brands present in those bases enter the models' memory with a firmer identity.

For most small and medium companies Wikipedia is out of reach in the short term, because it requires notability shown by independent coverage, and that does not block visibility. Wikidata sets a lower bar, though not zero (it asks for public references before the item), and entity consistency is built with the site, schema with sameAs, LinkedIn and the sector directories. Wikipedia comes when press coverage justifies it, and it should never be forced: promotional articles are deleted and burn the brand's credibility on the platform.

Yes, and for local brands they are among the strongest signals. When someone asks an assistant for a clinic, a restaurant or a nearby service, the engines lean on sources that carry social evaluation: Google Business Profile, review platforms, rated directories. A complete listing, with recent reviews and replies from the owner, is direct raw material for recommendations.

For digital and B2B brands the equivalent is the sector's review platforms and specialist directories, plus the public discussions where real customers talk about the product. The pattern is the same: engines prefer to recommend what independent voices can verify. Reviews are not bought or faked; they are asked for, systematically, from satisfied customers.

Because the engines look for genuine human experience, and forums like Reddit concentrate real discussions, with pros and cons, written by people who use the products. Several engines cite community platforms far out of proportion to their traffic, precisely because the content is less polished and more trustworthy than commercial pages.

The practical implication is that the public conversation about your category is feeding answers whether you take part or not. Taking part has rules: an identified presence, contributions with substance and zero spam, because communities expel and engines learn. One genuinely useful answer in a relevant thread can be cited for years; ten promotional ones damage the domain's reputation in those sources.

Yes, and in some respects it has the advantage. AI answers do not simply reproduce the Google ranking, and in many specific questions there is no dominant brand yet, because nobody built the quotable answer. Whoever gets to those questions first keeps them, regardless of size.

The small company's advantage is specificity: a well-defined niche, concrete questions from its buyer, depth the generalists do not have. Its disadvantage is a weaker entity, which is offset with consistency and patience. The realistic strategy is not to win the generic category questions but to own the specific ones where its knowledge is hard to match, and to grow from there.

For both, with different work. For digital brands (SaaS, e-commerce, online services) the game is played in extractable content, schema, topical authority and presence in the sources the models read. For local brands (clinics, restaurants, tourism, retail) the centre of gravity moves to a complete Google Business Profile, reviews with volume and recency, the same name, address and phone number in every directory, and clear geographic signals on the site.

Local questions to assistants are already routine: the best dentist in the area, a restaurant for a group dinner, who deals with a specific problem in the city. The engines answer from whatever local signals exist, and most nearby brands have not worked on any of them. It is one of the areas where the first to do the basics well gains fastest.