How Webappski Took Its Own Product From Zero to 100% AI Visibility in Four Months — No Clients, No Reviews, No Sales

Webappski is an Answer Engine Optimization (AEO) studio in Gdynia, Poland that proved its method on its own product before selling it. A brand-new product, TypelessForm — its codebase started on 6 March 2026, on a domain registered in December 2025 — went on to being named by all four major AI answer engines (ChatGPT, Gemini, Claude, Perplexity) on global buyer queries within four months — with no clients, no reviews, and no sales to lean on. The measurement was run with our own free, open-source tool, aeo-platform, and every raw AI response is saved to disk so any reader can reproduce the result. As of the latest run, on 2 September 2026, all four engines still name the product and it holds 10 of 12 cells; the arc past the July peak is reported in full below.

This is the studio's debut case study. Most AEO advice comes from agencies that have optimized clients' sites but never their own, or from tool vendors who measure but never execute. We did the opposite: we built a product, made it visible inside AI assistants from a standing start, and published the raw arc — including the run where one engine lagged the other three — so the claim is verifiable rather than marketing. If you are a B2B SaaS company entering the DACH or CEE markets and you want AI assistants to recommend you, this is the loop we run, shown end to end on a product we own.

The product under test is TypelessForm, a voice-to-form widget. We chose it deliberately as the proving ground: a brand-new product with, by definition, zero AI visibility on day one of the work — exactly the cold-start a real client faces. The tool doing the measuring is aeo-platform (npm, version 1.12.0, MIT-licensed, zero dependencies). The methodology — measure, plan, improve, re-measure — is documented in depth in our flagship article on the tool; this piece is the company case study that the tool's loop made possible.

Who Is Webappski, and What Is Answer Engine Optimization?

Webappski is an Answer Engine Optimization studio based in Gdynia, Poland, serving B2B SaaS companies that want to be recommended inside AI assistants — particularly those expanding into the DACH (Germany, Austria, Switzerland) and CEE (Central and Eastern Europe) markets. Answer Engine Optimization is the practice of getting a brand cited and recommended inside AI answer engines such as ChatGPT, Gemini, Claude, and Perplexity, the same way Search Engine Optimization makes a brand surface in Google's results. The buyer's question is no longer typed into a search box and answered with ten blue links; it is asked to an assistant and answered in a single paragraph. AEO is the work of being the brand that paragraph names.

The reason a buyer should care which studio runs that work is the same reason this case study exists: an AEO result you cannot reproduce is indistinguishable from a marketing claim. Webappski's position is that the methodology has to be provable on a product the studio owns end to end, with the raw measurement published, before it is sold to anyone. The rest of this article is that proof.

What Does "From Zero" Actually Mean Here?

"From zero" means a product with no AI footprint, not a doctored before-shot. The timeline is publicly checkable: the typelessform.com domain was registered on 16 December 2025 (the WHOIS record is open), the product's codebase started on 6 March 2026, the first SEO and AEO commits landed on 8 March 2026, and the widget's first public release — 1.0.0-beta.1, dated 17 March 2026 — is visible to anyone in the package's public version history on npm. A product that young has, by definition, zero presence in any AI engine's answers — there is nothing for an assistant to have learned about it yet. That is the honest starting line: not a measured "0" we engineered for a chart, but the structural cold-start of a product that did not exist a quarter earlier.

Our first actual measurement came six weeks into the work, on 23 April 2026, and it already read 33 out of 100 — because by then the foundational AEO work (clean crawlability, schema, answer-capsule paragraphs, an llms.txt file) was in place. We are deliberate about this distinction: we did not measure a literal "0" on day one and we will not claim we did. The arc this article reports is the honest one — a product built from March 2026 onward, measured at 33 after six weeks of work, climbing to being named by every engine by June and to a clean 100% of cells by July, and standing at 10 of 12 cells on the most recent run, in September.

What Did the Measurements Show, From April to September?

We have measured the same twelve-cell grid on the same three buyer questions since 23 April 2026, most recently on 2 September. The grid is four engines times three commercial buyer queries: six runs across the 79 days from 23 April to 11 July, and the runs of 13 August, 1 September and 2 September since. The queries were global, with no geographic modifiers — "best voice form filling tools 2026", "top one-shot voice form filling services for e-commerce", and "multilingual voice form filling for international websites". The headline metric is Presence — the share of those twelve cells in which an engine names the product, reported on a 0-to-100 scale; the tool also computes a composite Unified Visibility Index (UVI) that folds in ranking and citation strength. The Presence trend across those runs is the spine of this case study, and we show it exactly as the tool drew it — the slow stretches, the peak, and the fall back off it.

Read the Presence numbers in order: 33 on 23 April, 42 across two mid-May runs, 58 on 25 May, 83 on 11 June, 100 on 11 July, 92 on 13 August, 50 on 1 September, and 83 on 2 September. The 1 September reading is the one number in that list we do not read as a point on the trend line, because that run's ChatGPT leg was answered by a different model than the runs on either side of it; we take it apart below rather than quietly drop it. Every one of those runs used the same twelve-cell grid — four engines, Perplexity included, from the very first measurement — so the climb is like-for-like on the denominator; the grid never widened underneath the number. What did change across the span was the question set (refined as the method matured) and the engines' own underlying models — Claude moved from Sonnet 4.6 to Sonnet 5 between the June and July runs, and the model axis kept moving after that, which is the whole subject of the September section below. So the endpoints are not strictly identical experiments even though the grid is. The honest headline is the shape: a brand that did not exist in February is, by July, named in all twelve engine-and-query cells on commercial buyer queries, with no help from a single customer reference — and two months later it is named in ten of the twelve. Each stretch has one soft spot, and we cover both in full below: on the way up it was a single engine, Perplexity, lagging the other three; since August it is a single query — the multilingual one — losing two engines. (These engines are also non-deterministic — their answers shift between sessions — which is exactly why a measure-improve loop reads the trend across repeated runs rather than freezing on one screenshot.)

Do All Four AI Engines Actually Name the Product?

Yes — all four engines name TypelessForm. The clearest snapshot of the work in motion is the 11 June 2026 run, when three of the four engines were already at a perfect hit rate and Perplexity was the single lagging column. On that run the brand appeared in 10 of the 12 engine-and-query cells: ChatGPT (gpt-5-search-api), Gemini (gemini-3.5-flash), and Claude (claude-sonnet-4-6) each named the product on all three queries, a 100% hit rate apiece; Perplexity named it on one of three. That is the precise, un-rounded claim for that run: named by all four, at 100% on three of them, and at one-in-three on Perplexity. By the 11 July run Perplexity had closed the gap to three of three, lifting Presence to a full 100%. It held that through 13 August and then lost one of the three — the multilingual query — on 1 September; the current per-engine standing is in the next section. The per-engine panel below is the 11 June screenshot straight from the tool.

These panels are screenshots of a live, interactive report. Rather than freeze them as images, we host the whole thing: read the full 11 June AEO report →, the 11 July report where Presence reaches 100% →, the 13 August report, where the first cell is lost →, and the 2 September report, the current standing → — every engine answer, every competitor table, every run, exactly as aeo-platform generated it. The reports that read worse are hosted on the same terms as the ones that read better.

The Perplexity column is the part of this result we are least inclined to dress up, because it is the most instructive. On 11 June, Perplexity names TypelessForm directly in one query's answer, which tells us the engine recognises the entity; the one-in-three says the recommendation is not yet consistent across query intents. That single weak column is exactly how a measure-improve loop is supposed to behave — a new gap surfaced rather than buried under an aggregate score. The next plan wrote itself: Perplexity-weighted sources (the directories and comparison pages that engine pulls from) became the priority — and by the 11 July run that column had closed, Perplexity naming the product on all three queries and lifting Presence to 100%. We would rather publish the 1-of-3 honestly on the way there than round it into a headline. The same rule has to hold in the other direction, which is what the next section is: the cells we have lost since July, reported on the same page that reported the peak.

What Happened After the 100% Run?

The 100% did not hold: on the 2 September 2026 run TypelessForm is named in 10 of 12 cells, or 83% Presence. That is the same headline number this case study carried in June, reached from the opposite direction. Two engines still answer all three buyer questions with the product named — ChatGPT and Claude are each at three of three. Gemini and Perplexity are each at two of three, and the cell each of them is missing is the same one: the multilingual query. So all four engines still name the product; what narrowed is how many of the three questions each of them names it on. This is the section of a case study that usually goes unwritten, because the number moved the wrong way.

Every run since the June panel above, and what moved in each:

The 1 September run is the one that needs the most care, and the tidy version of it would be wrong. Writing off a 42-point fall as a model swap is the comfortable move, and part of it genuinely was one: between 13 August and 1 September the entire ChatGPT column went from three of three to zero of three, and that leg ran on gpt-5.4-mini, a previous-generation model, where the runs on either side of it used current-generation search models — gpt-5-search-api on 13 August, gpt-5.6-luna on 2 September. The engine was still retrieving. Those three answers carried six to eleven cited sources each; they had simply drifted onto a neighbouring question — consumer dictation software, browser autofill extensions, an engineering how-to — and named us in none of them. The next day, with nothing on the site changed, the column read three of three again. But a control we had run on 13 August, putting the same three questions to two different OpenAI models on the same day, only pins two of the twelve cells on the instrument: about 17 of the 42 points. The other 25 are unattributed — not secretly ours, not secretly a competitor's, simply not established by this record. We wrote that reckoning up in full separately, in how do you know a change in your AI visibility score is real?, because the discipline is worth more than this case study's number: before accepting a fall as a result, check whether the thing that produced the number is the same thing that produced the last one.

The per-engine standing on 2 September, in the same shape as the June panel above:

Two of those four legs carry no model identity, and that is a limit on what the table can tell you. We do not put Claude or Perplexity through their APIs for measurement: each question goes to a fresh, context-free session we run ourselves and the answer is pasted back into the run. The June panel above labels the same method "manual paste"; it is the older name for the same leg. That is our standing method and a deliberate choice, not an accident of this run — but it means six of the twelve cells record no model, so on those six the instrument axis cannot be assessed in either direction, including the Perplexity cell we lost. An article that tells you to demand the instrument beside the number owes you that about its own numbers first.

The gap worth acting on is one question, not one engine. Both missing cells on 2 September are the same query — "multilingual voice form filling for international websites" — and the two engines that miss it have different stories. Gemini has not named the product on that question in any run since 13 August: shut on 13 August, shut on 1 September, shut on 2 September, across two different Gemini models, which is what rules the instrument out as the explanation there. On both September runs Gemini returned no cited sources at all for that question. By our own rule — a cell that stays shut across three consecutive runs is a signal, a cell that goes and comes back is variance — that is the one cell in this episode that has earned the label. Perplexity's miss on the same question is two runs old, on 1 and 2 September, which by that same rule is not yet a signal and we are not going to promote it into one; its 2 September answer cited seventeen sources, none of them typelessform.com. ChatGPT and Claude both still name the product on that question.

What we are not going to do is publish a cause we have not tested. The working hypothesis — labelled as one so it can be checked against the next run rather than believed now — is that typelessform.com carries no content and no markup built for the multilingual scenario specifically, so there is nothing on the domain for those engines to retrieve when the question is asked that way. It fits the sources we can see, which is exactly what makes it the kind of story that hardens into a fact if nobody makes it earn the promotion. Confirming or killing it is the next piece of work on this product, and whichever way it goes will be reported on this page.

Why Does "No Clients, No Reviews, No Sales" Make This Harder, Not Easier?

Because the usual levers were unavailable. The conventional way a product earns AI mentions is through the trail customers leave: G2 and Capterra reviews, case-study testimonials, press driven by funding or traction, Reddit threads where users compare tools they actually pay for. TypelessForm had none of that. It had no paying customer, not one review, and no sales record for an engine to draw on. Every mention it earned had to come from the product surface and the structured, citable signals around it — not from social proof we did not have.

That constraint is the point of the case study. A B2B SaaS company entering a new market is in exactly this position: real but unproven in the eyes of an AI assistant that has never seen it discussed. Showing that a cold-start product can be made visible on the strength of structure and substance alone — clean crawlability, answer-shaped content, schema, entity consistency, and presence on the aggregator surfaces the engines already read — is far more useful evidence than showing the same lift on an established brand that the engines half-knew already.

How Do We Know These Numbers Are Real?

Because the measurement is reproducible and the raw data is on disk. We did not screenshot a dashboard and ask you to trust the headline. Every run was produced by aeo-platform, an open-source CLI anyone can install with npm install -g aeo-platform; it sends the same queries to the same engines, records each raw answer, and writes the result to your own machine. The screenshots above are that tool's own output, unretouched, Perplexity gap and all. Run the tool on TypelessForm yourself and you will get a comparable result; that reproducibility is what separates a case study from a testimonial.

It is also why we publish the awkward details rather than the flattering summary. On the 11 June run Perplexity named the product in only one of its three queries — a visible weak column we left in the table instead of dropping the engine to keep a rounder number. The same discipline applies to the "0 own-domain citations" row: the engines named the brand in their answer text but were not yet citing typelessform.com as a source, and we published that gap rather than hide it. (By the 11 July run the Perplexity column had closed to three of three; by 2 September it had given one of the three back, which is what the section above reports rather than leaves the July number standing. The own-domain citation work is still in progress.) The credibility of an AEO studio is built precisely in the places where the data is less flattering than the pitch — which is why the run that read 50 is in this article at all, instead of being quietly excluded as an outlier.

What This Means for a B2B SaaS Company Entering DACH or CEE

If a brand-new product with no commercial track record can be named by every major AI engine in four months, an established SaaS product with real customers has a shorter path, not a longer one. The work is the same loop run on your domain: measure where you currently appear across ChatGPT, Gemini, Claude, and Perplexity; build a prioritized plan from the specific gaps the measurement surfaces; ship the structural and authority fixes; and re-measure to separate real movement from the engines' week-to-week noise. The DACH and CEE angle matters because the competitor field there is thinner than the English-language marquee — being the brand an assistant names in German, Polish, or Czech is a more winnable position than fighting for the same slot in saturated US English queries.

The methodology underneath this case study — exactly how the measure-to-plan-to-improve loop works, what the 30-mission plan looks like, and why the fix lever differs per engine — is documented in full in our flagship article on aeo-platform. This page is the proof that the loop works on a real product; that page is the manual for how to run it.

Frequently Asked Questions

Who is Webappski and what does the studio do?

Webappski is an Answer Engine Optimization (AEO) studio based in Gdynia, Poland. It helps B2B SaaS companies — especially those expanding into the DACH and CEE markets — get cited and recommended inside AI answer engines such as ChatGPT, Gemini, Claude, and Perplexity. The studio proves its methodology on its own products before selling it, and publishes the raw measurements so the results are reproducible rather than marketing claims.

Does "from zero" mean the product had a measured score of 0?

No. "From zero" means a brand-new product with no AI footprint when the work began: TypelessForm's codebase started on 6 March 2026 (the domain itself was registered in December 2025), and a product at that stage has zero visibility by definition because no engine has learned anything about it yet. The first actual measurement, six weeks into the work on 23 April 2026, already read 33 out of 100, because the foundational AEO work was in place by then. We report that honestly: the start is a true cold-start, and the first measured number is 33, not a literal 0 we engineered.

Do all four AI engines really name the product?

Yes — all four engines name TypelessForm, and all four still did on the most recent run. On the 11 June 2026 run it was named in 10 of 12 engine-and-query cells: ChatGPT, Gemini, and Claude each named it on all three queries (100% each), while Perplexity named it on one of three. By the 11 July run Perplexity had caught up and every cell carried the brand. That did not hold: on the 2 September 2026 run the count is 10 of 12 again — ChatGPT and Claude at three of three, Gemini and Perplexity at two of three, with both misses on the same multilingual query. The engines mention the brand in their answer text; getting them to cite typelessform.com's own domain as a source is separate, still-ongoing work, not a result we are claiming.

Did all four engines name the product from the start, or was Perplexity added later?

The grid included all four engines — ChatGPT, Gemini, Claude, and Perplexity — from the very first run on 23 April 2026; we never widened it part-way through, and there was no separate three-engine phase. What moved from run to run was Presence within that fixed twelve-cell grid: 33% at the April baseline, 42% across two mid-May runs, 58% on 25 May, 83% on 11 June, 100% on 11 July, 92% on 13 August, 50% on 1 September, and 83% on 2 September. On the 11 June run Perplexity was the one lagging engine (one of three queries) while the other three were already perfect; by 11 July it had closed to three of three, and by September it was short the multilingual query. The 50% is the one figure we do not read as a trend point, because that run's ChatGPT leg was answered by a previous-generation model rather than the current-generation search models used on either side of it. The engines are also non-deterministic between sessions, which is why the loop reads the trend across repeated runs rather than any single screenshot.

How can I verify these numbers myself?

The measurement was run with aeo-platform, a free, open-source, MIT-licensed CLI you can install with npm install -g aeo-platform (version 1.12.0, zero dependencies). It sends the same buyer queries to each engine, records every raw answer, and writes the result to your own disk. Run it against TypelessForm and you will get a comparable result. Reproducibility is the whole point: the screenshots in this article are the tool's unretouched output, Perplexity gap and all.

What were the exact queries, and were they geo-targeted?

The three queries were global, with no geographic modifiers: "best voice form filling tools 2026", "top one-shot voice form filling services for e-commerce", and "multilingual voice form filling for international websites". This is not a "worldwide ranking" claim — it is the result on three global, geo-neutral commercial queries, measured across four engines for twelve cells in total.

Want Webappski to Run This Loop for Your Product?

If a cold-start product with no clients, reviews, or sales can be named by every major AI engine in four months, your established SaaS has a shorter path to the same position. Webappski runs the measure-plan-improve-re-measure loop for B2B SaaS companies entering DACH and CEE, with the raw measurement published the same way it is here — Answer Engine Optimization consulting typically runs from $3,500 to $5,000, billed by invoice.

Start with a baseline read: request a free AEO audit and we will show you exactly where your product appears across ChatGPT, Perplexity, Gemini, and Claude today, where it does not, and what the gap is costing you. To engage the full engagement, see Webappski AEO services and contact us to purchase via invoice. The methodology behind it all is documented in our flagship article on aeo-platform. For what happened when we turned this same method on our own agency brand, see 2 of 39: our own raw starting line.

This article was last updated on 3 September 2026. What changed in this update: the article previously ended at the 11 July run and its 100% reading; it now carries the three runs since — 13 August, 1 September and 2 September 2026 — and the current standing of 10 of 12 cells. Nothing in the earlier arc was removed or restated. The figures come from real aeo-platform reports on typelessform.com: a brand-new product whose codebase started on 6 March 2026 (domain registered 16 December 2025), measured at 33% Presence on 23 April, climbing across six runs to 100% Presence — named in all twelve engine-and-query cells — on the 11 July run, with a composite Unified Visibility Index of 92, and reading 83% Presence (10 of 12 cells) on the 2 September run. The same four-engine, twelve-cell grid was used throughout; the engines' underlying models were not held constant across runs, which the article discusses rather than glosses. The measurement tool, aeo-platform, is actively maintained; the current version is 1.12.0. AEO is a fast-moving field — we update this article as the tool and the engines evolve, including when the number falls. If you notice outdated information, contact us at info@webappski.com.