90 days ago we published the most uncomfortable post we have ever written. We are a digital agency that sells Answer Engine Optimisation. We help clients get cited in AI-generated answers. And when we asked ChatGPT about ourselves, it thought we might be a youth charity in San Francisco. This is a long one, so buckle up.
Our brand visibility in AI search was 0.21%. Our share of voice against competitors was 1%. Concentrate (who we hold in the highest respect) had 58%. We ranked for 8 keywords in Google, and every one of them included our own name.
We said we would fix it in public and share every number; good, bad and ugly. So here we go...
Search "best hubspot agency new zealand" from New Zealand today, and Google's AI Overview says this:
"Small-to-Medium Businesses (SMBs) & Onboarding: BLOOM Transitions is a Gold Partner known for CRM implementation, Shopify integration, and team training."
Named inside the answer and described accurately in our specific category. Search "hubspot audit nz" and we're there, too. Listed alongside two competitors as NZ specialists in our specific category.
Then try your search as a real live person would. A search of "who in new zealand can help with shopify and hubspot?" returns a list headed "Key New Zealand Providers" as the AI overview. Currently, we are first on it - ahead of a Shopify Premier Partner and another Elite Partner. We know AI search doesn’t stand still, but right now, we'll call it a win.
Across our tracked prompts, BLOOM now has an average visibility of 21.39% in AI-generated answers (as at 2 September 2026).
The number we did not expect is this one:
For the first time since we started measuring and tracking, BLOOM is at the top of our competitive set.
However, we do want to be careful about how much weight we put on that. This number is from one week of data in one tool, on one set of prompts that we chose ourselves. A larger, more established agency with a longer track record is still ahead of us on several measures. But nine weeks ago we were at 1% and they were at 58%. That gap has closed, and it closed because of specific work we can point to.
But look at us hanging out with the popular kids.
This is the clearest comparison we have. The same prompts, tracked in the same tool, from the first week to the last. No reframing, no cherry-picking.
Et voilà:
|
Prompt |
May |
Sep |
|---|---|---|
|
What agencies in NZ offer Shopify and HubSpot integration? |
1% |
97% |
|
Which HubSpot agencies specialise in SMBs in New Zealand? |
0% |
90% |
|
I need help setting up HubSpot for my business in NZ, who are the options? |
3% |
70% |
The first row is the one we are proudest of and is the most accurate description of what makes BLOOM different, and in May we appeared in 1% of answers to it. We now appear in 97%.
Our visibility score rose 15.34 points to 22.30%. Keywords in the top three went from seven to 14. Keywords in the top ten went from seven to 18.
Nine keywords now sit at position one, and eight of those now trigger a Google AI Overview:
Further down the list, "what is answer engine optimisation and which nz agencies do it" moved 94 places to position six. That was one of the prompts showing a dash (zero) when we started, and one we said would be the hardest to move.
It moved.
In the 30 July - 27 August AI citation landscape for our category, the most-cited URL was not HubSpot Marketplace. It was not a competitor. It was our own blog post. Shout out to our friends at Vanguard 86 for this one.
|
Rank |
URL |
Citations |
|---|---|---|
|
1 |
bloomtransitions.co/blog/the-6-best-hubspot-agencies-for-smbs-in-nz |
236 |
|
2 |
ecosystem.hubspot.com/marketplace/solutions/sales-and-marketing-alignment/new-zealand |
209 |
|
3 |
vanguard86.com/blog/new-zealands-top-hubspot-agencies |
171 |
That post is still climbing. In the most recent data we pulled, it reached 254 citations. (SEMrush 6 August - 3 September 2026).
Across all cited domains in that window, bloomtransitions.co sits second only to hubspot.com: 602 citations against HubSpot's 5,027. We're even ahead of salesforce.com at 579.
One well-structured, genuinely useful blog post outperformed everything else we published. It answers a real question, it names real agencies including our competitors, and it does not pretend BLOOM is the only option. That is why it gets cited.
So, to all the competitors we named in that blog: you're most welcome.
We promised honesty, so here is the ugly bit.
1. Our brand mention rate is still behind.
In the final week we were around 3 points behind our closest competitor. Being cited as a source is not the same as being named as a recommendation, and some are still named more often than we are. That's the next mountain and we'll be putting our boots on to climb it.
2. Our two tools disagreed wildly, and settling it taught us more than either number did.
HubSpot's AEO tool reports 19,408 citations across 2,040 responses. SEMrush's AI Visibility tool, pointed at the same domain and filtered to New Zealand, reports 1 citation, 1 cited page and zero mentions.
Rather than pick the flattering number, we tried to work out why, formed a theory, and then tested it. The theory was wrong, which turned out to be the useful part.
Some of the gap is straightforwardly definitional. SEMrush counts the number of responses that cite your domain. HubSpot counts every individual citation inside every response. Our HubSpot figure works out at around 9.5 citations per response, which is unremarkable given SEMrush's own published research shows ChatGPT averaging roughly 15 sources per response and Gemini around 3. One counts responses, the other counts citations within them. They were never going to match.
Some of it is coverage. HubSpot runs the 24 prompts we chose, weekly, across ChatGPT, Gemini and Perplexity. SEMrush auto-discovers its own prompt set and refreshes monthly. For our domain it surfaced just four topics and four prompts, one of which was "What is HubSpot CRM best used for in small businesses?" Four prompts is never going to agree with a 2,040 response sample. Its Distribution by LLM panel also returns n/a for ChatGPT, n/a for Gemini and n/a for AI Mode, so the two tools are barely looking at the same places.
And there is a wrinkle of our own making. We measure ourselves in SEMrush against New Zealand, because that is where BLOOM was founded and where most of our clients are. But our team sits across New Zealand, Australia and the United States, and we work with clients in all three. A single country filter was never going to capture that. When we opened the sample response for one of SEMrush's four prompts, with the New Zealand filter applied, the AI answer came back in Filipino. Make of that what you will, but it is a reminder that if your business does not fit neatly inside one country box, a country-filtered dashboard will not describe you accurately.
The only engine SEMrush returned a figure for was Google AI Overview, at 0%. That 0% is what we went to test.
Our SEMrush Position Tracking shows nine keywords at position one, eight of them flagged as triggering an AI Overview. SEMrush AI Visibility said we appeared inside none of them. Our working theory was that we ranked underneath the AI-generated box without being quoted inside it, which would have been a genuinely important distinction and one almost nobody draws.
So we opened an incognito window and ran the searches ourselves. We are inside the box. On both. Named, hyperlinked, and described accurately, in the exact category we sell into.
Which means the number that was wrong was the tool reporting zero. We do not know whether that is a New Zealand coverage gap, a sampling issue, or a definitional one, and we are not going to guess a second time. What we do know is that a dashboard told us we were invisible in AI Overviews while Google's AI was actively recommending us in them.
The lesson is not that one tool is bad. The lesson is that no dashboard is the territory. If a number matters to your business, go and look at the actual result with your own eyes before you report it to anyone. We had a plausible theory, sound reasoning behind it, and it took one incognito search to prove us wrong.
3. Our absolute traffic is still small.
SEMrush organic research shows bloomtransitions.co ranking for a handful of keywords with negligible estimated traffic. Visibility percentages are relative measures against a defined competitor set. They are not revenue. We would rather say that plainly than let a percentage do work it has not earned.
4. We only cover 6% of the possible prompt space.
HubSpot rates our prompt coverage at 6%. We are visible on the questions we chose to track. There is a much larger set of questions we have not yet touched.
5. We still have no presence on Reddit or YouTube.
Between 6 August and 3 September, LinkedIn generated 234 citations in our category and Reddit 130. We flagged this gap in month one and we have still not addressed it. It stays on the list until one of us puts their big girl pants on and films some YouTube content.
6. Client reviews took longer than expected.
We asked 3 clients for HubSpot Marketplace reviews in June. Chasing them politely, repeatedly, without being annoying, is genuinely harder than implementing schema. Side note: To our friends at Lentune, thank you, genuinely. Reviews feed directly into how AI answers questions like "who are the best HubSpot partners in New Zealand", which is one of the few measures where we are still behind. If you have worked with us and have not left a review, that is the single most useful thing you could do for us right now.
Alas, not yet. But we are eternal optimists and really good at what we do.
We set out to measure visibility, and visibility is what we measured. 90 days is a short window in agency sales cycles, and market conditions are genuinely tough right now. Anyone showing you a 90 day experiment with a tidy revenue figure attached is either working in a very different market to ours or being generous with attribution.
What we have built is an asset that compounds, content that gets cited and an entity that AI engines now describe correctly. A method we can run for clients because we ran it on ourselves first, in public, with the numbers published either way.
Whether that converts for BLOOM or not is next year's post.
Ranked by what the data shows, not by what we assumed at the start.
1. One genuinely useful piece of content.
Our most-cited asset by a wide margin is a blog post that honestly compares HubSpot agencies in New Zealand, competitors included. It earns its citations because it is useful, not because it is promotional.
2. Fixing the entity problem.
Organisation schema on the homepage, unambiguous copy on the About page, and consistent naming across every external profile. This resolved the San Francisco confusion within days and everything else was built on top of it.
3. Being where AI already looks.
Hubspot.com generated 5,027 citations in our category between 6 August - 3 September 2026. Rewriting our Solutions Marketplace listing put us inside the single most-cited domain in our space.
4. Peer relationships, not link building.
We reached out to Vanguard 86, another New Zealand HubSpot agency. They included us in their roundup, we agreed to refer work to each other where it falls outside the other's focus, and their pages generated 571 citations in our category between 6 August and 3 September.
5. Structured content and FAQ schema across service pages.
Slower to show results than the above, but the AEO services page now ranks and gets cited.
3 things we would tell you, based on doing it rather than reading about it.
1. The foundations move faster than anyone tells you.
We expected 8 to 12 weeks before anything shifted. Our disambiguation problem resolved within days of publishing Organisation schema. If AI is describing your business wrongly right now, that is often a fixable problem measured in days.
2. Being genuinely useful beats being promotional.
The single highest-performing asset we produced was a comparison post that recommends our competitors. AI engines cite content that answers the question, not content that sells.
3. You can't improve what you're not measuring, and you can't trust just one measurement.
We tracked weekly across HubSpot's AEO grader and SEMrush, with Google Search Console and Analytics for supporting traffic data. We checked the results by hand when the numbers did not agree. Half of what we learned came from noticing things we were not looking for, like the Gemini spike in the week where we did nothing. The other half came from checking a number we did not believe.
If you're curious about where to start, ask ChatGPT or Gemini two questions:
1. "What is [your business name] and what do they do?"
2. "Who are the best [your service] providers in [your region]?"
If the answer is wrong, or you are not in it, you have the same problem we had back in May. It is fixable, and it is more fixable than you think.
If you want to know where your brand stands, we can run your business an AEO audit. It covers how AI currently describes your business, whether you have a disambiguation problem, your citation profile, your schema, and a benchmark against your competitors. You get a prioritised action plan, not a fifty page report.
This experiment ends but the work doesn't.
Phase 2 is already scoped. We'll keep sharing the numbers, trials and tribulations - just a bit less often for now.
Thank you to everyone who followed along, supported us on our journey, challenged our methodology on LinkedIn, or messaged to say they had asked ChatGPT about their own business and did not like the answer. That last group is the reason we did this.