from page four
to the answer box.
Hello Customer is a voice of customer platform from Belgium. In the summer of 2026 I repositioned the brand, rebuilt the whole website in three languages with search engines and AI answer engines in the room, and then cleaned up everything around it. This is what the first seven weeks look like, with the data. The new site went live on 18 July 2026 and every number here was exported on 4 September, so read it as an early reading, not a final one. SEO compounds; this is the start of the curve.
ask an ai about voice of customer platforms. hello customer is in the answer.
Four screenshots from the first week of September 2026, unedited. The same questions in May came back with a different set of names.
"best alternatives to qualtrics." Listed in the enterprise CX section beside Medallia, and the third source card is Hello Customer's own Qualtrics alternatives article, published on 3 August. The comparison page did exactly the job it was built for.
"what is a good nps." The definition sentence is sourced "Hello Customer +4". This is one of the highest-volume questions in the category, and the site's rewritten NPS cluster is the reference.
"how to choose voice of customer platform." The opening line of the answer is cited to Hello Customer, first source card. The category we repositioned into, answered by the site we built for it.
"what is best voice of customer platform." In the shortlist with Qualtrics, Medallia, Enterpret and Chattermill, tagged "European mid-market" with "a strong balance of collection, AI analysis and action". In the July baseline, unbranded English prompts mentioned Hello Customer zero times out of fifteen.
in may, google thought the site was about its customers' logos.
When I opened Search Console, the terms bringing the site the most impressions were luminus, colora and securitas. Not the product. The logos of three customers, sitting on a customers page that happened to rank first for them. Nearly a quarter of all impressions came from people looking for a company that wasn't Hello Customer.
The site's title was "Business critical insights through feedback". It's a decent sentence. Nobody types it into anything, and no model files it under a category.
The average position across everything Google showed the site for was 38. Page four. In AI Overviews, the site earned around 200 impressions a day, almost all of them in the Netherlands and Belgium. The US and the UK barely registered. And in the baseline I ran in July, unbranded English prompts to ChatGPT mentioned Hello Customer zero times out of fifteen.
So the problem wasn't "not enough content". The site had 280 blog posts. The problem was that nothing on it said, in the words a buyer or a model would use, what the company is.
of all impressions were people searching for a customer's name or logo. Struck-through rows are that traffic. Bars are relative; hover for the share.
reposition first. then rebuild everything, with the crawler and the model in the room.
I don't think you can bolt search onto a finished site. The positioning decides the keywords, the keywords decide the page architecture, and the architecture decides what a model will ever cite. So the order was: category, then pages, then everything behind the pages.
one intent per page, named the way people search
Product pages were named after features ("CX Signals", "Forward-looking alerts"). Buyers search for "customer feedback alerts" and "CX reporting". Every page got one job and a title that a stranger would type.
the category in the title, everywhere
"Voice of customer platform" in the homepage title, the hubs, the schema and the llms.txt. If you want to be filed under a category, say the category.
be the comparison
Buyers ask "alternatives to Qualtrics" and models answer with lists. So the site now has the honest list, for 14 categories, in three languages. Comparison pages went from 9% to 22% of all AI impressions.
answer the question on the page
FAQ sections with FAQPage schema on every product page. BlogPosting plus FAQPage on every article. And an NPS calculator built into the article that explains NPS, so the page does the thing it describes.
entities, not just keywords
The site already had schema. It had three overlapping blocks per page, a LinkedIn sameAs pointing at a company that no longer exists, and no FAQ on ten product pages. Fixed on 75 templates.
retire ruthlessly
158 legacy posts, 63 old case-study landing pages, ebooks, events and a press section: drafted and redirected to the surviving page on the same subject. Less for the crawler to waste a budget on, and the model reads a site that agrees with itself.
speak the language properly
French and Dutch built from native glossaries mined from the client's own older writing, not translated. Then reviewed line by line: nearly 4,000 fixes, including one listicle that claimed a competitor's product was end-of-life. It wasn't.
make it fast and easy to read by machines
A render-blocking script loaded on every page and used by none was removed. llms.txt, IndexNow, hreflang with x-default, self-referencing canonicals, a sitemap with nothing stale in it.
Positioning. New shell, 15 product pages, homepage with interactive Listen, Analyse and Act sections.
French and Dutch variants of every page. Go-live on the production URLs, 18 July.
Blog refresh, schema overhaul, legacy sweep, Core Web Vitals, llms.txt.
Comparison project, CX article series, full French and Dutch language reviews, IndexNow.
Measure. Everything on this page is from the 4 September exports.
we gave up a quarter of the traffic on purpose. the rest got better.
Search Console, the same 28 days a year apart: August 2025 against August 2026. Percentages, because the point isn't the size of the site, it's the shape of what changed.
The customer-logo searches went from nearly a quarter of impressions to under one percent. That traffic was never going to buy anything, so I let it go, and it's the reason the total impression count looks flat while everything inside it moved.
What replaced it: the site now earns impressions on 57% more pages, sits on page one for 2.2 times as many queries, and clicks from non-brand searches are up by about a third. The "voice of customer" family of queries doubled in count, grew 78% in impressions, and moved from page six to page three on average.
The average position across everything is the number I watch most, because it's the one that can't be gamed by adding pages. It went from 38 to 18. And keep the calendar in mind: the 2026 window starts eighteen days after launch, so this is Google's first read of the new site, not its settled opinion.
One result in ten gets clicked on page one. Almost nothing gets clicked on page four. Moving the average from 38 to 18 is the difference between existing and being seen.
three times the ai visibility, and it stopped being a benelux story.
Google's generative AI report only started reporting in mid-May 2026, so the earliest clean baseline is 18 May to 1 June, seven weeks before launch. Against 5 August to 1 September, per day, because the windows are different lengths.
Impressions in AI Overviews and AI Mode went from about 200 a day to about 600. The number of pages being pulled into AI answers grew from 198 to 315, and the number of countries from 101 to 152.
The bigger change is where. In May, three quarters of AI impressions came from the Netherlands and Belgium. In August the United States was the joint-largest market, at more than seven times its May level, and English pages went from 69 impressions a day to 358. The comparison pages alone moved from 9% to 22% of everything.
Same site, same product, seven weeks in. What changed is that the pages now say what the company is in the words the model was already using. I'll re-measure the same panel each quarter; the interesting number is the one after this one.
cited in one answer out of four
Bing's report doesn't give a before, so this one is a snapshot. Across 146 buyer questions where Copilot grounded its answer on hellocustomer.com, the site was cited 5,122 times in the window, a citation-weighted share of 24.5%. On 65 of those questions it's in at least a quarter of the answers. Type a word to filter, or click a column to sort.
| question copilot was answering | citations | share of answers |
|---|
what i'd tell you, if this were your site
positioning is a search decision
The category name in the title did more than any single page. Before you write anything, pick the word buyers and models already use for what you are, and then say it, plainly, in the places machines read first.
one intent per page, and mean it
Feature names squatting on keywords, three schema blocks disagreeing with each other, ten product pages with no FAQ. None of it is glamorous to fix. All of it is what a crawler and a model actually see.
be the comparison
Buyers ask "alternatives to X" and models answer with a list. Write the honest list yourself, in every language you sell in. The comparison pages went from a tenth of AI visibility to nearly a quarter in six weeks.
retire more than you add
158 posts and 63 old case-study pages redirected. Less for the crawler to waste a budget on, and the model reads a site that agrees with itself. Cleaning up did as much as building.
I did this hands-on: the positioning, the pages, the schema, the redirects, the calculator inside the NPS article. With Claude, extensively, which is how one person rebuilds a site in three languages in seven weeks and then reviews four thousand lines of French and Dutch. And I'll say it once more because it matters: this is seven weeks of data. It's enough to show the direction, not the destination. If you'd like the same done to yours, the button is below.
what is aeo, answer engine optimisation?
how long did the results take?
where do the numbers come from?
did organic traffic go up?
want your site
in the answer?
Book a call. We look at what Google and the models currently think your site is about, and whether that's what you'd like them to think.
