You know your business has changed. Your positioning is sharper, your services have evolved, and you know exactly who you help and how. But when a potential customer asks AI to recommend a business like yours, will it describe the business you are today, or the one you used to be? It might even leave you off the list entirely. And the reason may have less to do with your website than you think.
We asked five AI tools about our own business
I recently asked five AI tools two questions about our business.
First: Describe Whippet Digital in two sentences.
Second: Which agencies in New Zealand specialise in SaaS growth and AI search?
I used a fresh chat in each one, so no conversation history could influence the answers.
Four of them described Whippet as a New Zealand SaaS growth agency built around an eight-pillar framework, with AI search at its core.
That is who we are today.
The fifth called us a “full-service digital marketing agency” offering SEO, content, advertising, social media and websites.
That is who we used to be.
Same business. Same question. Two very different descriptions.
What came back
The four accurate answers came from ChatGPT, Gemini, Perplexity and Google. Each picked up similar themes from our website: SaaS growth, AI search optimisation and a structured framework that connects visibility to pipeline. Several echoed our own wording.
When I asked which New Zealand agencies specialise in SaaS growth and AI search, those tools included Whippet. ChatGPT called us the strongest fit for both.
The fifth answer came from Claude. It described us as a general digital agency and cited a single third-party directory listing as its source. It didn’t mention our website at all.
When I asked Claude the buyer question, Whippet didn’t appear.
If a SaaS founder had asked that tool for an agency, we might never have made their shortlist. We would never have known.
These were our results on that occasion, not a definitive test of how every AI tool behaves. But they were enough to make us look more closely.
Why the answers differed
AI doesn’t know your business the way you do. It builds its picture of your business from information it can access, retrieve or has learned.
When those sources agree, your story is easier to recognise. When they conflict, the picture can become muddled, incomplete or simply out of date.
That’s what happened to us.
Our website now describes us as a SaaS growth specialist, but some third-party directories still describe a general agency, with different founding dates and services.
One tool went with the directory.
That tool wasn’t necessarily being careless. It was drawing on information it found and considered relevant. The problem was that the information didn’t reflect who we are today.
This matters because AI is becoming part of how buyers research and shortlist potential suppliers.
G2’s 2026 research found that 51% of B2B software buyers surveyed started their research with an AI chatbot more often than with Google. The survey included 1,076 buyers and was conducted in March 2026.
If AI doesn’t understand what you do, who you help or why you’re a good fit, you may never make it onto the shortlist.
And you may not even know you’ve been overlooked.
Being recommended, consistently and accurately, is becoming an important part of being discovered.
The uncomfortable lesson?
You can change your business, your website and your positioning, and AI may still describe the old version of you until the information it draws on catches up.
What you know that AI doesn’t
There’s a difference between what AI can find out about your business and what you know about it.
Priyanka Vergadia’s TED talk, What You Know That AI Doesn’t, made me think about this.
You know what you’ve learned from clients, how your offering has evolved, what makes you different and where you create real value.
AI can’t reliably infer all of that from scattered or outdated information.
Your experience and judgment are valuable. But if the evidence of that expertise isn’t visible where AI can access it, it may not make it into the answer.
That’s the gap to close.
And it’s not just about writing more content. It’s about making sure the right information exists, is accurate and is discoverable.
Three things to do this week
You don’t need a complicated AI visibility strategy to get started. Begin with these three things.
1. Write one clear description of your business.
Agree on a sentence or two that explains what you do, who you help and the problem you solve.
Use it as your reference point across your website, LinkedIn page, directory listings and other profiles.
The wording doesn’t have to be identical everywhere, but the facts and positioning should tell the same story.
2. Find and fix the old versions.
Search for your business name and look beyond the first page of results.
Check directories, listings, partner pages, social profiles and old website content.
Look at the basics:
- What you do and who you serve
- Your services and specialisations
- Where you’re based
- Your founding date and business history
- Your team size, if published
- Any former business names
Update what you control. For the rest, request a correction or claim the profile where possible.
It’s slow, unglamorous work. It’s also where our experiment found a problem.
3. Run the two-sentence test yourself.
Open ChatGPT, Gemini, Perplexity, Claude and Google in fresh chats. Use a consistent set of questions.
Start with:
Describe our business in two sentences.
Then ask a buyer-intent question, such as:
Which companies in [your market] specialise in [your service or solution]?
Write down what each tool says, word for word. Record the date, the tool and any sources cited or linked.
Check three things:
- Is the description accurate?
- Does it reflect your current positioning?
- Are you included when a buyer asks for recommendations?
Where practical, repeat the prompts. AI answers can vary, so look for recurring patterns rather than judging everything on one response.
Then fix what you can and run the same test again in a few weeks.
Tools update at different speeds, and changes to public information won’t necessarily show up immediately or consistently.
What we’re doing next
We’ve started fixing the inconsistencies we found, updating our website footer and several third-party listings, including G2, Zyppy and DesignRush.
But updating a source and changing an AI answer aren’t the same thing.
Some changes may be picked up relatively quickly; others may take longer. We can’t assume every tool will reflect them at the same time.
So I asked Claude how long it might take to recognise our changes.
How long before AI catches up?
The answer highlighted something important: there isn’t just one route for new information to reach AI, and timelines can vary widely.
1. AI search: potentially days to weeks, but no guarantee
Anthropic operates separate crawlers for different purposes:
- Claude-SearchBot helps improve Claude’s search results.
- Claude-User can retrieve web pages when someone asks Claude a question.
- ClaudeBot collects public web content that may be used in model training.
If someone asks Claude a question that triggers web search, it may find updated information before the model’s built-in knowledge reflects it.
Anthropic doesn’t provide a guaranteed recrawl or update schedule, so we can’t put a firm timeline on when our changes will appear in search-grounded answers.
If someone asks Claude to look directly at whippetdigital.com, it may retrieve the updated page, provided access is available and the content isn’t being served from an outdated cache.
There’s another complication. When I tried to retrieve our G2 listing, it rejected access. If a page can’t be accessed directly, Claude may need to rely on search indexes or other sources mentioning it.
2. AI’s built-in knowledge: potentially months or longer
This is a different story.
Anthropic’s ClaudeBot collects public web content that may contribute to future model training. Updating a website or directory doesn’t mean that information will immediately become part of Claude’s built-in knowledge.
That depends on future training and model updates, and we can’t promise a reliable timeframe.
What we’re checking now
There are two practical things on our list:
- Check our robots.txt settings to make sure we’re not unintentionally restricting the relevant Anthropic crawlers.
- Establish a baseline using a few specific prompts, such as “AI search agency for B2B SaaS in New Zealand”, then repeat those tests over time.
Anthropic explains what its crawlers do and how website owners can manage access in its Claude Help Centre.
The takeaway? Correcting your public information is a worthwhile first step. But don’t assume that because you’ve updated it, every AI tool will immediately tell the new story.
The part AI can’t do for you
This experiment was meant to be a quick check. It turned into a reminder that being good at what you do and being described correctly are two separate jobs.
AI can help people discover you. But it can’t make your business story accurate if its information is outdated or inconsistent.
You have to do that part.
AI will summarise whatever it finds.
You decide what information you share.
What AI does with it isn’t entirely in your control.
Try the two-sentence test
Ask an AI tool to describe your business, then ask it to recommend businesses like yours.
Getting the description right is only the start
A consistent story across the web is one of the eight pillars in our SaaS Growth Framework. The other seven focus on being found, trusted, and chosen by buyers and AI alike.
Does the answer reflect who you are today?
What surprised you? I’d be interested to hear.
