How perfume lab The Elixir doubled its Google traffic and got named in AI Overviews and ChatGPT answers
Context and task
The Elixir Perfume Laboratory is a perfume lab with two locations: Kyiv (Gulliver mall) and Ivano-Frankivsk (Promprylad). It runs perfume-making workshops (solo, for two, group and on-site) and sells gift certificates, reed diffusers, car fragrances and custom scents for brands.
At the start, on 4 August 2026, traffic came mostly from the brand name and the homepage. In 30 days the service pages had 10 clicks and 544 impressions. For non-brand workshop and certificate queries Google showed the homepage, not the page where a visitor could pick a format and book.
The second part of the task was AI answers. At the strategy session we saw that models describe the business as a lab and a workshop, while the site was betting on "bespoke perfume". On top of that, answers about the brand kept repeating the wrong number of years in business.
The bet
A separate page for each service format works twice. Classic search finds it for a specific query, and AI Overview takes facts from it for the answer: address, format, duration, bottle size, price. If those facts are written briefly and identically on the site and off it, the lab is easy to name in an answer next to older workshops.
What we did
Classic SEO stayed the base, and every content decision was checked against one question: can a machine take a self-contained passage from this page?
- Technical audit and a brief for the developer: redirects, canonical URLs, indexing.
- New service structure: a services hub and a workshops hub, separate pages for solo, couple, group and on-site formats, gift certificates and sets, a "Perfume making" branch. 26 pages in total, built from our briefs and texts; the client's team published them.
- Answer first: addresses, formats, duration, bottle sizes and prices as short statements at the top of each page.
- Experience signals: "About us" and "Our team" pages, a hub of the lab's own cases.
- Shop and blog: descriptions for six categories, updated metadata, 11 articles on gifts and expertise, a Russian-language version of the pages.
- Brand facts: we checked what ChatGPT, Gemini, Perplexity and AI Overview say about The Elixir and removed the wrong years in business from the site.
- Structured data: LocalBusiness with both locations on the homepage.
- External publications on topical sites about workshops, personal scents and gifts.
Google Search Console, service, certificate and perfume-making pages: 268 clicks and 7,999 impressions in 30 days versus 10 and 544 before the start (5 Jul–3 Aug vs 23 Sep–22 Oct 2026).
Results in Google Search
We compare the 30 days before work began (5 Jul–3 Aug 2026) with the latest 30 days with data (23 Sep–22 Oct 2026).
- Clicks to service, certificate and perfume-making pages: 10 → 268
- Impressions of those pages: 544 → 7,999
- Clicks, whole site: 589 → 1,286
- Impressions, whole site: 12,368 → 24,944
- Average position, site: 8.6 → 6.3
- Non-brand clicks*: 85 → 308; brand clicks*: 108 → 99
- Queries in TOP-10 / TOP-3 of 163 (SE Ranking, 1 Sep → 5 Oct 2026): 67 / 25 → 101 / 49
* Only queries that GSC reports: 193 of 589 clicks before the start and 407 of 1,286 after. Brand queries contain "elixir" or "еліксир" in any spelling.
Non-brand clicks moved the most: they tripled even without the "Scent of Donetsk" queries, while brand clicks stayed flat. The growth came from people looking for the service, not for the lab's name.
"Scent of Donetsk" is the client's own news story: 121 clicks to that case page and 52 non-brand clicks on queries about it. Without that page the site had 1,165 clicks, almost twice as many as before the start.
AI Overviews and ChatGPT answers
We separate two states: the brand in the list of sources under the answer, and the brand named in the text a person reads. Below is the second.
- Google AI Overviews, 21 and 24 Oct 2026: 4 queries, brand in the text in 3 of 4; in two of those three answers the site is among the sources.
- ChatGPT, 26 Oct 2026: 2 questions, The Elixir first with a direct recommendation in 2 of 2.
- ChatGPT, 5 Nov 2026: 5 questions, first with a direct recommendation in 5 of 5.
Conditions: Ukrainian language, Ukrainian market, signed out; in ChatGPT a new chat for each question. There were no technical failures. Link URLs in ChatGPT answers were not saved, so domain citation is not counted there. Each screenshot is an example answer, not a stable position. The screenshots are in Ukrainian, the language the queries were asked in.
Google AI Overview, «сертифікат на створення парфумів» (perfume-making gift certificate), 24 Oct 2026. The first sentence of the answer cites The Elixir site, and the lab opens the "Popular options and prices" block.
Google AI Overview, «парфумерний майстер клас івано-франківськ» (perfume workshop Ivano-Frankivsk), 21 Oct 2026. The lab is named in the first sentence, and the client site is among the sources. On the repeat check the wording changed, the brand stayed.
Google AI Overview, «створення парфумів київ» (perfume making Kyiv), 21 Oct 2026. The Elixir in the list of places with the Gulliver mall address and formats. The model took the price from an Instagram post published before the price increase, so facts outside the site need aligning too.
The open gap and ChatGPT
For «майстер клас парфуми київ» (perfume workshop Kyiv) The Elixir is not in Google's answer: Be Craft, Le Flacon and art studios are.
In ChatGPT the first snapshot was on 26 October, the repeat with a wider set on 5 November. Five questions in plain language, the way a gift buyer asks them, each in a new chat. In all five answers The Elixir comes first, alongside Be Craft, KatArtSys, Ol.factory and other workshops. Three of the five answers are below.
ChatGPT, "Perfume-making workshop in Kyiv: where to go?", 5 Nov 2026. For a query where AI Overviews does not show the brand yet, ChatGPT opens with The Elixir, labelled "My favourite".
ChatGPT, "Where in Kyiv can I make my own perfume and buy a gift certificate?", 5 Nov 2026. The Elixir first, "My favourite for a gift", the same result as on 26 October.
ChatGPT, "Where to hold a hen party or birthday with a perfume workshop in Kyiv?", 5 Nov 2026. "My favourite for a hen party" and a link to the group workshop page created within the project.
Limitations
No AI baseline. AI Overviews and ChatGPT answers were not recorded before 4 August, so we do not claim the brand appeared in them because of our work.
Personalisation. Answers change between loads, regions and sessions. The screenshots are examples, not a full list of queries.
Old prices. All three AI Overviews answers show prices from before the October increase: the model takes them from third-party sources.
ChatGPT: five questions, two days. These are snapshots, not regular measurement; Perplexity and Gemini were not measured. ChatGPT does not bring traffic yet: 72 sessions from chatgpt.com in June and 14 for 1–19 October (GA4). Being named in an answer and getting clicks from it are different things.
Attribution. The growth comes from new pages, content, technical fixes and the client's own news together; we do not attribute it to one factor.
Conclusion
The lab had demand before we started. What it lacked were pages Google could send that demand to. Once every service format got its own page with the facts up front, those pages earned more clicks in a few weeks than in the whole previous period, and AI Overviews and ChatGPT cite them in their answers.
The same answers show the limit: the model also collects facts outside the site, from Instagram, event listings and directories. So the next stage is to align prices and formats everywhere Google looks for confirmation, and to win the «майстер клас парфуми київ» query.
Data: Google Search Console (sc-domain:theelixir.ai), 5 Jul–3 Aug vs 23 Sep–22 Oct 2026; SE Ranking, 163 queries, Google Ukraine; live Google AI Overview results in Ukrainian, 21 and 24 Oct 2026; ChatGPT signed out, 26 Oct and 5 Nov 2026; GA4. Case prepared by Dmytro Popryadukhin, Head of SEO.
