Frequently asked questions

Question 1 of 12

Why GeoVouch — and what is it for?

GeoVouch is an entity-level GEO evaluation product for teams responsible for how a specific product, brand, or sellable offer is discovered by AI. It is built for product marketers, growth and SEO/GEO teams, founders, and ecommerce operators who own an official product homepage, product landing page, or store listing. People increasingly ask an AI assistant for a recommendation instead of scanning a page of search results. When they do, the assistant answers from a handful of sources it decided to trust, and everything else is simply absent from the answer. Being absent is not the same as ranking tenth: there is no second page to be on. Traditional SEO tools tell you where you rank on a results page, which is a different question from whether an assistant names you at all.

1. Product GEO Analysis starts with a product entity: a recognizable name and an official page, domain, or listing that the evidence can be attributed to. It evaluates that entity's own pages and the public conversation around it — not just the technical quality of an arbitrary URL.
2. Assistants recommend from what they were trained on and what they retrieve — round-ups, comparison articles, review sites — far more often than from a product's own marketing pages.
3. Whether they can read your site at all is a technical question with concrete answers: robots rules, structured data, whether your claims are stated in a form a model can extract.
4. Whether independent sources corroborate what you say about yourself is a different question again, and usually the harder one to fix.

GeoVouch measures those things separately, with the evidence attached, so you can tell which of them is actually holding a product back. It is not intended to score a news or information aggregator, directory, or open community/UGC destination as though it were one product: those sites represent many unrelated entities and need a publisher- or platform-level methodology instead. If a product appears within one of those properties, analyze that product's official page or identifiable store listing. GeoVouch does not promise to change an assistant's answer — it shows you what that answer is currently built on.

↑ Back to top

Question 2 of 12

How does GeoVouch work?

Both workflows produce a dated snapshot of observable evidence. Nothing is inferred from a proprietary index: every score points at the pages, search results, or model answers it came from, and you can open them.

1. Product GEO Analysis starts from one URL. It identifies the product and its official homepage, audits that page for crawlability and structured data, then reads the product's own content, then what independent third parties publish about it.
2. It then switches to a de-branded view — searching the way someone who does not know your brand would — to see whether you surface in category results at all, and how much of the category's comparison coverage is about you.
3. Prompt Result Analysis skips all of that. You give it one question, it asks several models that question twice each, and it ranks which products those answers actually named, recommended, and cited.
4. Every step stores its result, so revisiting a finished step costs nothing and shows exactly what it produced.

Steps are charged individually, as they run, and each one tells you its price before it starts. A run you stop halfway is only charged for the steps that finished.

↑ Back to top

Question 3 of 12

What does GeoVouch measure?

GeoVouch measures observable AI-discovery signals rather than claiming a universal AI ranking. Product GEO Analysis starts from a website and separates the product's own technical and content signals from category search visibility and independent third-party evidence. Prompt Result Analysis starts from a buyer-style question and records what selected models actually answer and cite.

The result is a dated evidence snapshot. It is useful for prioritizing work and tracking changes, but it is not a guarantee that a search engine or AI assistant will always recommend a product.

↑ Back to top

Question 4 of 12

When should I use Product GEO Analysis versus Prompt Result Analysis?

Use Product GEO Analysis when you want to understand one product's full readiness: whether its site is technically readable, whether its official content is clear, and whether it appears beyond its own domain. Use Prompt Result Analysis when you want to test a specific question a buyer might ask and compare which products models name today.

The two views complement each other. A product-level report explains the conditions around discoverability; a prompt result shows the current outcome for one concrete buyer question.

↑ Back to top

Question 5 of 12

Why can the same prompt or website produce different results later?

The web, search results, model routing, and model behavior all change. A search-grounded answer may use a different source set on a later run, and the content of an analyzed website may have changed as well.

1. Use the dated response rows and cited URLs to inspect what changed in a particular run.
2. Use Prompt result stats for the current completed result of tasks that used the same exact prompt.
3. Use Model citation stats to see cumulative source patterns for a model's usable answers within the selected time range.

A rerun replaces that task's current result; it is not presented as a permanent archive of every intermediate model response.

↑ Back to top

Question 6 of 12

What counts as a citation in Model citation stats?

A citation is a source URL associated with a usable model answer in a completed Prompt Result Analysis task. The statistics aggregate the selected model's cited domains and URLs across completed tasks in the selected time range.

If search retrieval returns URLs but the model produces no usable answer body, GeoVouch shows those URLs as diagnostic search sources but excludes them from citation counts and Model citation stats.

↑ Back to top

Question 7 of 12

Which models can I test, and how are they run?

The model picker exposes text models available through OpenRouter. The default list emphasizes currently popular models; searching a model family surfaces newer matching models first when that metadata is available.

Each selected model runs the same prompt independently. Availability, source grounding, latency, price, and answer quality remain properties of the selected model and its provider, and can change over time.

↑ Back to top

Question 8 of 12

What are Credits, and how are they used?

Credits are GeoVouch's simple unit for complete analysis work. They cover the paid AI-model, search, and web-retrieval services required to produce a result; GeoVouch does not expose provider token counts or individual API costs to you.

1. New accounts receive 20 welcome Credits. Each UTC calendar day, signed-in users can claim 1 free daily visit Credit from the Credits menu; claiming is available once per day.
2. Credits pay for the AI, search, and web-retrieval services an analysis actually uses. Product GEO Analysis is charged step by step as each one runs — Official Content, 3rd-Party Content and Category Search Results cost 1 Credit each, Share of Voice costs 2 because it searches and reads considerably more, and the Final Report is free because it only summarises steps you already paid for. A complete run is 5 Credits; stopping early costs only what finished. Prompt Result Analysis costs 1 Credit per model. Revisiting a step that already ran is always free, and a step that fails returns its Credits.
3. A Prompt Result Analysis costs 1 Credit per selected model. One Credit includes that model's two independent responses, citation processing, product recognition, and result statistics.
4. Preparing an analysis, reading saved results, and retries caused by a GeoVouch or provider failure do not use additional Credits. Starting a fresh completed analysis does use Credits again.

Credits are shared across Product and Prompt analyses, purchased once, and never renew automatically. Your balance is checked and reserved before GeoVouch starts a chargeable analysis, and a reservation is released if the step fails. The Terms of Use cover how Credits are charged and refunded.

↑ Back to top

Question 9 of 12

What data does GeoVouch retain?

GeoVouch needs to retain the information required to provide an analysis: account email and session records when signed in, URLs and prompts you submit, selected models and settings, task status, results, citations, and timestamps. Primary storage is Supabase (managed PostgreSQL in AWS us-east-2), with per-account and per-visitor isolation enforced by database row-level security; OpenRouter provides model access.

One deliberate exception to per-account isolation: a prompt you analyze, and the products and URLs models named in response, are added to a shared corpus that every user's analysis draws on. Those records carry no account or visitor identifier, but the prompt itself is visible to others — treat a prompt as public. The Data Privacy page describes retention, third-party processing, and deletion in full. Do not submit secrets or sensitive personal information in a prompt or target URL.

↑ Back to top

Question 10 of 12

What are the trade publications, industry blogs, and YouTube channels that AI chatbots cite?

There's no single fixed list — it's whichever independent publications, blogs, and creators already cover your specific category, and that shifts over time as new outlets and creators show up. Four concrete ways to find the ones that actually matter for your product, roughly in order of how directly they connect to what this analysis measures:

1. Look at what this analysis already found. The domains listed under Category Share of Voice and Category Search Results in your report are publications already ranking and getting cited for your category's searches — that's a stronger, more current signal than any generic list could be.
2. Ask an AI chatbot directly. Search ChatGPT, Claude, or Perplexity for "best [your category] tools" and note which sources it cites in the answer — those are literally the sources feeding AI answer engines' recommendations right now.
3. Search for active review coverage. A Google search for "[your category]" review OR blog OR youtube surfaces the publications and creators currently covering the space.
4. Check review-aggregator "learn" sections. G2, Capterra, and similar platforms publish their own category guides and roundups, and often link out to press coverage worth reaching as well.

Corroboration depth and mention quality specifically need independent, third-party discussion — not the product's own site — so the goal of all four methods is the same: find real, currently-active independent voices covering your category, not a fixed "trade publications" list that may not even cover software.

↑ Back to top

Question 11 of 12

What is a "known-cited source", and why do sites like Reddit, Wikipedia, YouTube, and G2 count more than others?

A known-cited source is a domain that's independently documented as one AI answer engines (ChatGPT, Perplexity, Google AI Overviews) frequently pull from when they cite sources:

1. Community — Reddit, Stack Overflow.
2. Review — G2, Capterra, TrustRadius.
3. Reference — Wikipedia.
4. Press — major outlets.
5. Developer — GitHub docs.
6. Video — YouTube.

Corroboration depth and mention quality both weight these more heavily than an evidence item from a domain outside that list: a handful of substantive mentions on a few known-cited platforms scores higher than a larger pile of mentions from unrecognized, low-authority sites. The scoring isn't just counting how much independent coverage exists — it's weighing how likely that coverage is to actually feed what a chatbot ends up citing back to a user. This is also why the corroboration and mention-quality actions in Improvement Suggestions point you at G2, Capterra, and TrustRadius by name for reviews, and at trade publications, industry blogs, and YouTube channels more broadly for press coverage — both are ways of getting your product discussed on the kind of source this scoring already knows AI answer engines lean on.

↑ Back to top

Question 12 of 12

I fixed something myself, but a rerun still shows the same problem — did the fix not work?

It depends on which dimension you fixed. This report is built from two fundamentally different kinds of checks, and only one of them can confirm a fix the moment you make it:

1. Internal GEO dimensions — Technical GEO (F1-F6) and Official Content Analysis — crawl your own site directly, every time you rerun them. A fix here (a meta tag, a robots.txt line, a dateModified field) shows up on the very next rerun, because the check is reading your live page, not a copy of it.
2. External GEO dimensions — Category Search Results, Category Share of Voice, Corroboration depth, and Mention quality — don't crawl your site at all. They read live Google search results for your category and brand. A correct fix here (a new page, a review request, a Wikidata entry) first has to be found, crawled, and indexed by Google — or, for Wikidata, reconciled into Google's Knowledge Graph — before it can show up in a result these checks can find. That can take anywhere from a few days to several weeks, entirely outside this tool's control and outside yours.
3. Every action in Improvement Suggestions is labeled with which of these applies — "Verifies on your next rerun" for internal fixes, "May take time to verify" for external ones — so you know which kind of wait to expect before rerunning.

A same-day rerun showing no change on an External GEO action almost never means the fix was wrong — it usually just means Google (or Wikidata/Google's Knowledge Graph) hasn't caught up yet. To sanity-check progress without waiting on a full rerun, search Google directly for your new page's exact title, or search site:yourdomain.com, to see whether it's been indexed at all yet.

↑ Back to top