The Honest Truth About AI Visibility
More Articles
.jpg)
Over the last few months, almost every client conversation we’ve had eventually gets to some version of the same question: "Do we show up when people ask AI about us?"
It's a fair question, and one we struggle to answer because we're not prepared to sell anyone an expensive dashboard that promises accuracy but is dictated by wildly uncontrollable variables.
This post is our honest answer: why wanting AI visibility is a legitimate goal, and why most of the tools currently being sold to you are shakier than they look.
We’ve drawn on existing reporting and testing done by others in the SEO and marketing space, linked throughout. We're synthesising it because we think our clients deserve the honest version of this story before they sign a contract, not after.
Wanting to show up in AI answers is a real goal
People genuinely are using ChatGPT, Gemini, and Claude (or any LLM in fact) the way they used to use Google - asking for recommendations, comparisons, and "best of" lists. Only this time the prompts are more conversational and almost never the same (the first tricky part!). Users benefit from getting one clear answer, rather than clicking through the top ten blue links and perhaps getting frustrated with sponsored options (ads) which are not relevant to the query.
Industry benchmarks are starting to reflect that shift, though it's worth noting most of these numbers still come from the same vendors selling the tracking tools, so take any specific conversion-rate stat with a pinch of salt. The underlying behaviour change, though, isn't in dispute - if your customers are asking AI assistants who the best option is in your category, you want a reasonable shot at being mentioned. That part of the pitch is sound - and you’re not wrong for wanting it for your business!
Where it goes wrong
A lot of AI visibility platforms describe four trackable dimensions: whether you appear, how prominently, how favourably, and which sources the AI cited. On paper it reads like rank tracking with extra steps. In practice, several structural problems undermine that comparison.
The prompts aren't representative. These tools work by running a fixed set of sample prompts on a schedule and checking whether your brand shows up (this brings about another issue we'll mention later). What they don’t guarantee is that users are searching for these prompts. There’s no tool like Google Keyword Planner which gives you an estimate of monthly searches.
Research done by SparkToro asked real users for the prompts they'd be likely to type in about a category, and found barely any overlap between them. Across 142 survey responses, barely two of the prompts collected even looked similar to one another. People don't reduce intent to two or three clean keywords the way they do with Google, they ask specific questions based on their situation. A tool reporting "you appeared in 14% of tracked prompts" may be measuring 14% of a list nobody asked.
Unlike a search ranking, an LLM's response isn't a fixed position you can poll reliably. In recent years, search ranking itself has gotten harder to track for the same underlying reason: every SERP (Search Engine Results Page) is personalised to the person searching. The usual fix, checking rankings in an incognito window, gives you an unbiased baseline, but if your actual customers aren't searching in that way, the baseline stops reflecting what they see.
LLMs take that same problem and turn it up. Gemini's Personal Intelligence feature went into beta in January 2026 for Google AI Pro and Ultra subscribers, and connects to a user's Gmail, Workspace apps, Photos, YouTube and Search history to shape answers without being asked. ChatGPT builds a persistent memory of saved facts and past chats, and OpenAI raised its custom instructions limit from 1,500 to 5,000 characters in July 2026 alone, giving every user an even longer standing brief that shapes the tone of every answer and which brands get named.
The same Sparktoro research, found that when testing across nearly 3,000 data points of identical questions, asked repeatedly, returned completely different brand recommendations from one run to the next. ChatGPT and Google AI both showed less than 1% chance of producing identical brand lists in any two runs out of 100. Claude performed slightly better at around 2-3% consistency, but that’s still not a number you want to be hinging marketing or strategic decisions on. Most tracking tools run their checks logged-out with memory and personalisation switched off, which again gives you a clean baseline, but it's a baseline that fewer and fewer real users experience.
It's like asking a big group of people "pick a number between 1 and 100." You’ll get very few duplicates on the same try. Ask the same crowd the same question tomorrow and you'll get a completely different spread - not because anyone changed their mind, but because there was never a fixed answer to begin with. Each response is a fresh roll of the dice.
That's closer to what's happening: the AI isn't recalling a settled opinion, it's generating a fresh pick each time from a range of plausible options. When tracking tools are running prompts to determine where your brand sits, Search Engine Journal has flagged a version of the observer effect here that can greatly affect the data you’re seeing both in these platforms and in analytics such as GA4. The automated polling these trackers do can generate traffic and fetch patterns that skew the very analytics being used to judge whether a strategy is working. In a post from Jan-Willem Bobbink, an SEO consultant, shared on X, he says, “When you run daily prompt tracking…the LLM sometimes decides it needs fresh information to answer. It fires off a RAG request and your pages get crawled. Those crawls land in your server logs as bot hits. They inflate your crawl data. They dirty your content performance analysis.” So it could be that the only traffic seeing your brand, is the traffic generated by the tracking tools you’ve paid to judge performance on. Basically, paying for a service to help you do better, and the data it’s using to analyse and advise you on, is skewed by its own actions – like a football coach running on the pitch and kicking the ball for you…
If you’re savvy enough to want to determine where the LLM is getting data about your brand from, you’ll be looking at the citations. The issue with this is citations aren't consistently visible or logged, even to the platforms. Some engines show source links in the interface but don't store or expose them for public verification, which makes trend data over time hard to independently check. That’s not to mention any hallucinating the LLM may do (a very real thing!), that’s typically an issue with longer running conversions, but it’s still not to be discounted. Much like people, LLMs aren’t always telling the truth. It’s not their fault though, they’re at the behest of their modelling and the inputs they’re receiving. Actually, that’s arguably just like humans who lie!
None of this means the tools are useless, they’re just not the Godsend that you’re being sold. Directional trend data is meaningful when looking at if you’re gaining or losing ground relative to competitors over months. Treating a single score as a precise, client-reportable metric the way we'd report a Google ranking is not something the current generation of tools can honestly support.
Stop asking what AI says about your business
Talking about AI sounds like you're on top of the problem, addressing the newest gadget, wanting to know how your business comes across. As the section above lays out, that's close to impossible to do properly. You can't ask every version of every question a real customer might type, and the answers aren't stable even when you do.
LLMs aren't forming a verdict of their own. They're gathering what's already being said about you across the internet and piecing it into one neatly packaged reply. So, if you're wondering what an LLM has to say about you, the real question is: what do your customers already say about you?
AI has given business owners a false idea that there's a single thing to fix. There isn't, the answer's bigger than that. LLMs are reflecting your own brand perception: what's already being said about you on social media, in forum threads, wherever people compare products before they buy. If there's a flaw in your messaging or a gap your product doesn't cover, yes, an LLM might surface it. But the LLM isn't what needs fixing.
We all need to stop looking for an easy fix and a single tool to blame. In fact, don’t we just need to fix what the tools are looking for?
What is worth doing?
While concentrating your LLM visibility down to a single number is not something we recommend, being conscious of how your brand might appear is a valid part of your marketing strategy. The good news is most of what helps you show up in AI answers is the same work that helps you show up everywhere else.
- Structured, clearly organised content. Proper headings, FAQ sections, and schema markup make it easier for any system be that a search engine or LLM, to lift accurate information about you.
- Being cited by other credible sources. LLMs draw heavily on the broader web, not just your own site. Digital PR, industry mentions, reviews, and third-party coverage build the same kind of authority signal that's always mattered for SEO, and it's one of the more defensible levers for AI visibility too.
- Answering the real questions your customers ask, in plain, factual language, rather than obsessing over the specific handful of prompts a dashboard happens to track.
- Basic technical health. Crawlable, fast, accessible pages that both search engines and AI crawlers can read.
- Genuine expertise and specificity on your own site, rather than generic category pages, since specific, well-sourced claims are what tends to get picked up and repeated accurately.
None of this requires a specialised AI-visibility subscription. It's SEO fundamentals, done properly, with an awareness that a wider range of systems are now reading your content.
Where we land
If a client asks us to help them show up more in AI search, the honest answer is yes, and here's how, and it looks a lot like good SEO with a slightly wider lens. Simple really? Like “SEO+” if that’s easier to understand. What we won't do is put a client's budget into a premium dashboard that reports a "visibility score" with false confidence, when the industry's own testing suggests that score can swing wildly between identical runs. We'd rather tell you that plainly now than have you find out later.
So, if you’re thinking “I need to up my marketing game with AI search, AEO, GEO, etc etc” then you’re not wrong, but the first place to look is at what already exists in your marketing ecosystem. Is that correct, and already doing a good job? If so, it can be tweaked and improved to add this into your mix, subtly. If not, then the SEO work needs addressing deeply, with the AI lens on quick – we can help you with both!
Curious about SEO and AEO? Get in touch
