LLM Perception Drift: Why Your AI Visibility Changes Month to Month

A business that ChatGPT recommends this month may disappear from its answers next month, even if nothing obvious has changed.

Your rankings may be stable. Your website may be the same. There may be no penalty, technical problem, or sudden drop in traffic. The AI simply isn’t mentioning you as often as it did before.

So why does that happen?

What is LLM perception drift?

LLM perception drift is the change over time in how AI models understand, reference, and recommend a brand.

Jordan Koene of Previsible introduced the term to describe month-over-month changes in the way AI models position brands within a category.

That’s an important distinction from traditional SEO.

With Google rankings, we became accustomed to thinking in positions. You ranked third for a keyword last month and fourth this month. Something moved, and usually we could investigate why.

AI recommendations don’t behave quite like that.

A business can appear in one response and disappear from the next. The sources being cited can change. Competitors can suddenly start appearing more frequently. Even the way the model describes your business can shift.

That means AI visibility isn’t really a position you win and hold. It’s better thought of as something you keep earning.

Why can AI visibility change when nothing changed on my website?

Because your website is only one of the signals influencing what an AI model says about your business.

The wider information environment is constantly changing.

Competitors publish new content. Reviews come in. News articles get published. Directories are updated. New discussions appear on Reddit and other communities. Search indexes change. AI platforms change the sources they rely on.

Your website can remain completely unchanged while the world around it changes significantly.

That’s one of the biggest adjustments businesses need to make when thinking about AI search.

If your ChatGPT visibility drops, looking only at your website may not tell you why.

How does an AI model actually know about my business?

Broadly speaking, an AI assistant can know about your business through information learned during training and through information it retrieves when answering a question.

Those two sources of information move at very different speeds.

Think of them as two clocks.

What is the slow clock?

The slow clock is the information and associations learned during model training.

Once a model has been trained, that underlying knowledge doesn’t simply update every time something changes on your website.

If an assistant answers a question about businesses or brands without retrieving current information, its response can be influenced by what it learned during training.

This creates an obvious limitation.

You might receive fantastic press coverage this week, publish a major piece of research, or completely reposition your company. That doesn’t mean the underlying model immediately learns it.

Those signals have to become part of the broader information environment and, eventually, may influence future models.

That’s why building a clear, widely recognized brand entity is a long-term job.

What is the fast clock?

The fast clock is retrieval: the current information an AI system can find and use while generating an answer.

This layer can change much faster.

A competitor publishes a useful guide. Your Google Business Profile changes. A prominent article mentions your company. Your reviews increase. An important directory updates its information.

Any of those things can change what information is available to an AI system when it searches.

That’s why some changes in AI visibility can happen surprisingly quickly.

The practical lesson is that these two layers shouldn’t be treated the same way.

Fresh content, updated pages that have structured content for AI search engines, and wider distribution can influence what gets retrieved relatively quickly. Building stronger associations between your company, your expertise, and your category generally takes much longer.

How volatile are AI recommendations?

Very volatile, which is why checking a prompt once tells you surprisingly little.

Kevin Indig, working with AirOps, analyzed 815,000 prompt-page pairs. When the same prompt was run three times in ChatGPT, only 2.2% of citations remained across all three runs.

His conclusion was memorable:

“Reporting a point estimate from one run is astrology.”

The point isn’t that AI recommendations are completely random. They aren’t.

The point is that one answer is a very small sample.

If you ask ChatGPT for the best medspas in Toronto this morning and your business appears, you haven’t suddenly “won ChatGPT.”

If you run the same prompt tomorrow and disappear, you haven’t necessarily lost anything either.

What matters is frequency.

How often does your business appear across repeated runs?

How does that frequency change over several weeks or months?

That’s a much more useful measurement of AI visibility than a screenshot of one favourable answer.

What can cause my AI visibility to drop?

Usually it’s not one dramatic event. It’s a change in the collection of signals surrounding your business and your competitors.

Several things can contribute.

A competitor may start publishing heavily around an important topic and become more closely associated with it.

Their review profile may improve while yours stays flat.

Your strongest pages may become dated while newer sources enter the search results.

A directory or third-party website may contain outdated information about your business.

A competitor may receive significant press coverage or mentions from authoritative sources.

Or the AI platform itself may simply change how it retrieves and weighs different sources.

That last one is especially important.

You don’t control ChatGPT, Gemini, Perplexity, or Google’s AI systems. The sources they favour today won’t necessarily be the sources they favour six months from now.

This is why building AI visibility around one website or one third-party platform is risky.

Can AI visibility improve just as quickly?

Yes. The same forces that can work against you can also work in your favour.

If you’re publishing useful information, earning credible mentions, improving your reviews, and becoming more visible across the web, including by getting featured in Google AI Overviews, AI systems have more reasons to associate your business with the topics you want to be known for.

This is an important way to think about perception drift.Drift isn’t inherently negative.

Your competitors are drifting too.

The opportunity is to make sure the overall information environment increasingly supports your business rather than theirs.

How do I know what kind of drift I’m experiencing?

Start by separating changes in retrieved answers from broader changes in how consistently AI systems associate your brand with a category.

This isn’t always perfectly clean because modern AI products can combine several systems when producing an answer. But it is still a useful diagnostic.

Run your important prompts repeatedly and record whether your business appears, which sources are cited, and which competitors are mentioned.

Then look for patterns.

If your visibility is dropping primarily in answers built from current web sources, investigate the retrieval environment first.

What sources are replacing yours?

Which competitors are being cited?

Have newer pages appeared?

Are competitors earning mentions from sources you don’t have?

Are your important pages still current?

Those are problems you can often start addressing immediately.

If your business continues to be retrieved and cited but appears less frequently in broader unaided brand recommendations, you’re dealing with a longer-term brand and entity problem.

That usually requires sustained work rather than one quick optimization.

How should I measure AI visibility properly?

Measure repeated appearances across a fixed group of prompts rather than checking random questions whenever you remember.

You don’t necessarily need an expensive platform to start.

A spreadsheet is enough.

The important part is using a consistent methodology.

Which prompts should I track?

Start with 20 to 30 questions that closely resemble the questions your actual customers would ask.

Don’t include your company name.

For a medspa, that might include:

“Best medspa in Oakville”

“Where should I get Botox in Oakville?”

“Best clinic for laser hair removal near Oakville”

For a personal injury lawyer:

“Best personal injury lawyer in London Ontario”

“Who are the top car accident lawyers in London?”

“What lawyer should I call after a car accident in London Ontario?”

Once you’ve chosen the prompts, keep them as consistent as possible.

If you change half the questions every month, you’re changing the measurement at the same time you’re trying to measure the result.

How many times should I run each prompt?

Ideally, run important prompts several times rather than relying on a single answer.

Three to five runs is a reasonable starting point.

The AirOps research illustrates why. Identical prompts can produce very different citation sets across repeated runs.

So don’t record only whether you appeared.

Record how often you appeared.

If your business appears in four out of five runs one month and one out of five the next, that’s much more useful information than saying you were “ranking third in ChatGPT.”

Which AI platforms should I monitor?

Monitor the platforms your customers are likely to use, because visibility on one does not guarantee visibility on another.

At minimum, we would look at:

• ChatGPT
• Google AI Overviews
• Google AI Mode
• Gemini
• Perplexity
  

Don’t combine all of them into one vague “AI ranking.”

They retrieve information differently, use different systems, and can produce very different recommendations for the same question.

Even Google’s AI Overviews and AI Mode should be treated as separate surfaces.

What numbers should I actually track?

Four measurements give you a useful starting point: appearance frequency, share of voice, mention rate, and position.

Appearance frequency: In what percentage of your test runs did your business appear?

Share of voice: How often did you appear compared with the competitors you’re tracking?

Mention rate: Was your business actually named in the answer, or was your website simply used as a source?

Position: When you were recommended, where did you tend to appear within the response?

The first measurement is particularly important.

AI visibility is probabilistic. Thinking in terms of frequency usually gives you a much better picture than trying to force AI recommendations into traditional ranking reports.

How do I know whether a change is real or just noise?

Look for repeated movement across multiple runs and over time rather than reacting to individual answers.

If your business disappears from one ChatGPT response today, don’t rewrite your website tomorrow.

That’s noise.

If your appearance frequency falls from 70% to 55%, then 40%, then 25% across a meaningful sample of prompts and repeated runs, that’s something worth investigating.

This is where AI visibility reporting needs to be different from traditional rank tracking.

The trend matters more than the screenshot.

How often should AI visibility be audited?

For most businesses, a monthly benchmark with more frequent checks on high-value prompts is a sensible starting point.

We generally want enough time between full audits to see meaningful movement without waiting so long that a problem goes unnoticed.

Important commercial prompts can be spot-checked more frequently.

Then every quarter, step back and look at the competitive picture.

Who is appearing more frequently?

Which sources are supporting them?

Which topics are they becoming associated with?

Where are they earning mentions that you aren’t?

That’s often where the most useful information appears.

How do you make AI visibility more resilient?

Give AI systems multiple consistent reasons to associate your business with the services, expertise, and locations you want to be known for.

There is no single GEO trick that accomplishes this.

The strongest businesses tend to build visibility from several directions at once.

Should I publish more content?

Yes, but consistency and usefulness matter more than publishing huge amounts at once.

A burst of 30 articles followed by six months of silence isn’t a particularly strong strategy.

Publish steadily around the subjects where you have genuine expertise.

More importantly, make the content useful enough that people and other websites have a reason to reference it.

Do third-party mentions matter?

Yes, because what other websites say about your business can provide evidence that your own website cannot.

Your website can say you’re one of the leading providers in your market.

That’s expected.

It’s much more meaningful when respected industry publications, local media, professional organizations, podcasts, directories, and other independent sources consistently connect your business with that expertise.

This is one reason digital PR and traditional SEO are increasingly overlapping with AI search optimization.

Does business information need to be consistent everywhere?

Absolutely. AI systems have a much easier job understanding a business when multiple trusted sources agree about it.

Your business name, location, phone number, services, hours, and other important details should be consistent across your website, Google Business Profile, and relevant directories.

This matters particularly for local businesses.

SOCi’s 2026 research also shows just how difficult AI local visibility remains: in its dataset of more than 350,000 locations, recommendation rates varied substantially across ChatGPT, Gemini, and Perplexity.

When several sources disagree about basic facts, you’re giving an AI system uncertainty it doesn’t need.

Should evergreen content be updated?

Yes, when there is something meaningful to improve.

Update statistics. Add new information. Answer questions that weren’t covered previously. Improve weak explanations. Add examples. Remove outdated sections through content pruning.

And while you’re there, make sure the page is structured clearly enough that both a human reader and an AI system can quickly understand the answer.

How important are reviews?

Reviews remain an important part of the wider reputation signals around local businesses.

Don’t think only about your average rating.

A healthy review profile is ongoing. Customers continue leaving reviews, the business responds appropriately, and the profile reflects the current customer experience rather than what people were saying three years ago.

For local businesses in particular, reviews are part of the broader information environment AI systems can encounter when researching recommendations.

Should I focus heavily on Reddit or another platform that AI systems cite?

Use important third-party platforms, but don’t build your entire AI strategy around one of them.

Citation patterns can change.

A source that receives significant exposure from an AI platform today may receive less tomorrow.

The safer strategy is diversification.

Build your own authoritative content. Earn press. Develop industry mentions. Maintain strong business profiles. Encourage genuine customer reviews. Participate in relevant communities.

You want multiple paths leading back to the same conclusion about your business.

What is the biggest lesson from LLM perception drift?

AI visibility isn’t something you achieve once and then forget about.

There is no permanent state where your company is “optimized for ChatGPT.”

The models change. Retrieval systems change. Competitors change. Sources change. Your market changes.

So the goal isn’t to chase every individual fluctuation.

It’s to build enough consistent evidence around your business that you’re repeatedly
associated with the subjects, services, and locations that matter to your customers.

Then measure whether that association is getting stronger or weaker.

That’s what makes LLM perception drift useful.

Instead of asking:

“Does ChatGPT recommend us?”

Ask:

“How often are AI platforms recommending us, how is that changing, and what sources appear to be influencing those answers?”

That’s a question you can actually measure.

And it’s the difference between discovering a visibility problem when a client points it out and catching it while the movement is still small. Connect with Logik Digital and our team will do a  website audit and trace the reasons why your business faces this drift.

Frequently Asked Questions:

1.Does paid advertising improve AI visibility?

Not directly in the same way that publishing a useful source or earning a credible mention can.

Buying ads doesn’t simply buy your way into ChatGPT recommendations.

Paid campaigns can still contribute indirectly by increasing brand awareness, searches, visits, and discussion around a company.

But we wouldn’t treat advertising spend itself as an AI citation strategy.

2. If I stop publishing, how quickly will my AI visibility fall?

There isn’t a reliable universal timeline, and anyone giving you an exact number is probably oversimplifying it.

Different industries, prompts, and AI platforms behave differently.

A highly established brand with years of coverage and third-party mentions isn’t going to behave like a small local business whose visibility depends on three recent articles.

What we can say is that retrieval-based AI visibility can change quickly because the sources available to the model are constantly changing.

That’s why monitoring matters.

You want to notice a sustained decline while it’s still a trend you can investigate, not six months later when somebody asks why competitors suddenly seem to be everywhere.

3. Can a few negative reviews cause a sudden drop?

Possibly, but there is no established number of negative reviews that automatically causes an AI platform to stop recommending a business.

Reviews are one signal among many.

The overall rating, number of reviews, recency, responses, the content of those reviews, and the other information available about the business can all matter.

So rather than trying to find a magic threshold, watch the trend.

If your review profile is weakening while competitors are consistently earning strong recent reviews, that’s worth addressing whether or not you can prove it caused a particular ChatGPT answer.

4. What should a small business spend on AI visibility monitoring?

You can start for essentially nothing.

Create a spreadsheet.

Choose 20 to 30 commercially important prompts.

Run them consistently.

Record which businesses appear, which sources are cited, and how frequently your business is mentioned.

Paid monitoring tools start becoming worthwhile when you’re tracking enough prompts, competitors, locations, and AI platforms that doing the work manually takes too much time.

The methodology matters more than the software.

5. Why did one of my competitors suddenly start appearing everywhere?

Usually because something changed in the information environment around them, although you need to investigate before assuming you know the cause.

They may have published a strong cluster of content.

They may have received media coverage.

Their review profile may have improved.

An authoritative website may have added them to an industry list.

They may have strengthened their presence across several sources at once.

Or the AI platform may simply have changed which sources it retrieves.

The useful question isn’t, “How did they game ChatGPT?”

It’s:

What evidence is the AI finding about them that it isn’t finding about us?

That question gives you something you can actually work with.

Earn Your Place in the Answer 

At Logik Digital, an AI SEO Agency in Toronto, we build the prompt set, establish the baseline, and monitor AI visibility over time across the platforms that matter. That gives businesses a clearer picture of where they’re appearing, where competitors are gaining ground, and where the underlying signals need strengthening.