GPT-6.1 Terra: what is this model and what does it mean for brands?
GPT-6.1 Terra is OpenAI’s mid-sized model, released on 26 October 2026, designed to run on a computer: it understands what…
A decline in AI performance: a temporary blip or a genuine problem?
A drop in AI visibility in a single measurement is usually just a fluctuation, as AI chatbots answer the same question slightly differently each time. A genuine problem is indicated by a decline that persists across several consecutive measurements using the same set of questions, affects a single model following an update, or is accompanied by a loss of page citations. Before you take any action, check whether the model, the calculation method or the set of questions has changed.
- How much fluctuation in visibility is normal in AI?
- How does a fluctuation differ from a fall?
- What should you check first when your results go down?
- How do model updates affect the results?
- When should you react straight away, and when should you wait?
- How do you show a customer a decrease?
- How does Aelo help you tell the difference between a fluctuation and a decline?
Below, we explain how much fluctuation is normal, what to check first, and when to take immediate action.
How much fluctuation in visibility is normal in AI?
More than one might intuitively expect from SEO. In a study by SparkToro and Gumshoe.ai, ChatGPT and Google’s AI provided the same list of brands in fewer than 1 in 100 repetitions of the same question. Even leading brands in the headphones category appeared in 55–77% of the responses, but not in every one. The model generates text with an element of randomness; on the next run, it may search different sources, and the result also depends on the account and the model version. The mechanism describes the text Why does AI give a different answer every time?.
That is why a single result is not a data point, but a sample. What matters is the percentage of responses associated with the brand – that is the key metric brand visibility in AI, calculated from a number of measurements taken at the same frozen query set, calculated separately for each model. The website explains how often to take measurements to ensure this percentage remains stable How often should visibility be measured?.
How does a fluctuation differ from a fall?
Stability and scope. A fluctuation appears in one measurement and disappears in the next. A drop in visibility in the AI persists across several measurements or affects multiple queries at once. The third possibility is an event: a sudden change in a single model on a specific day, usually following an update.
CHANGE TO THE RESULT
A fluctuation, a trend or an event
Every situation calls for a different response.
CASE 1
FluctuationThe score drops in one assessment, then returns to normal in the next. The change relates to individual questions.WHAT TO DONothing. Make a note of it and compare it with the next reading.
CASE 2requires a response
Downward trendThe score drops across several consecutive measurements or across many questions at once.WHAT TO DOCheck the citations, competition and website using the list below.
CASE 3
EventA sudden change to one model from a specific date; no changes to the others.WHAT TO DOCheck whether the model has received an update, and separate the periods in the report.
Always compare the same set of questions and each model individually.
aeloapp.ioThe average across all models can mask each of these scenarios. A brand may lose ground in one chat and gain ground in another, whilst the average remains unchanged. That is why the report shows the results separately for each model, as we explain in the text AI Visibility Report for the client.
What should you check first when your results go down?
First, look at what might have affected the measurement itself, and only then at what might have affected the brand’s position. The order matters, because you can check the first three points in a few minutes, and they often explain the entire decline.
AFTER THE FALL
What to check and in what order
The first two points concern measurement; the following ones concern brands.
01Set of questions and methodAre the questions the same as before, and has the way the score is calculated changed?
02Model updateDoes the drop affect a single chat and begin on the day its version is updated?
03Website citationsHas the model stopped referencing your website in questions where it used to do so?
04Competition in the answersHas a new competitor emerged in your brand’s market, or has a new source appeared that the model is now using?
05The website and robotswhether the key subpages are working, have not changed their URLs and are not blocked in the robots.txt file for AI search engine bots
The highlighted point usually indicates a genuine decline rather than a fluctuation.
aeloapp.ioIf the decline is due to lost citations, look for the reason in the content: the model may have found more specific details elsewhere, such as the price, specifications or standards. If a new competitor has appeared in the responses, check which sources the model is using to obtain information about it. The text will help Why does AI recommend the competition? and a description of the mechanism in the text How AI chatbots choose the brands they recommend.
How do model updates affect the results?
They are able to shift the results from one day to the next, without any change on the part of the brand. An example from October 2026: on 7 and 8 October, ChatGPT switched to GPT-6, and free and paid users were given two different models. Results from before and after that date must be compared as separate periods. Details are set out in the text GPT-6 in ChatGPT: what it means for brand visibility.
Changing the calculation method in the tool works in a similar way. In the case of the lime manufacturer, the visibility metric stood at 63% in July and 44% in September, whilst at the same time the number of queries mentioning the brand rose from 17 to 18 out of 18, and the number of page citations rose from 50% to 94% phrases. Between these measurements, the method of calculating the indicator changed, so the reports do not compare these two figures. The sequence of events is described in the following sections of the case study: An AI model recommends a brand for the first time and Citing a webpage in AI. The method of calculation in Aelo is described in methodology.
When should you react straight away, and when should you wait?
Immediately, if the drop is down to you: a subpage that isn’t working, a changed address without a redirect, a robot block, or incorrect company details in the new source. In the latter case, the text will help AI is providing false information about the company.
You should also take action when a drop in ratings indicates a change in the brand’s status: a model who previously recommended it now merely recognises it, or has stopped mentioning it even when asked directly. The first scenario is described in the text Knows the brand, but doesn’t recommend it, the other side ChatGPT doesn’t know my company, and the actions that reinstate the recommendation, text How to get AI to recommend your company.
Wait – if a drop appears in a single measurement, it relates to individual questions and is not accompanied by a change in citations or competition. Changes to content and sources take time to take effect, and hasty adjustments following every poor result make it difficult to assess what actually worked. The website explains why the effects take weeks to materialise Why do changes take time?.
How do you show a customer a decrease?
As part of a trend, not as a single figure. In the report, show the results of several measurements, separately for each model, and highlight any events that may have influenced them: model updates, changes to methodology, and actions implemented. This way, the client can see whether the decline is a fluctuation, a trend or a specific event before they even ask. It is worth standardising the terminology: reference, citation and sentiment mean the same thing in every report.
How does Aelo help you tell the difference between a fluctuation and a decline?
A decline in visibility within AI can only be assessed through a series of measurements. We measure whether ChatGPT, Claude, Gemini and Perplexity mention your brand, using the same set of questions every day, and we show the results separately for each model, including citations and mentions of competitors in the responses. The measurements are carried out by the module Visibility, and when the drop has a cause that needs to be rectified, the task is assigned to Tickets. The text describes the whole process How Aelo works, and you’ll take the first measurement on Free account.
Agencies that explain these fluctuations to their clients can find out more on the website Aelo for agencies.
FAQ
Is a drop in AI visibility in a single measurement a cause for concern?
Usually not. AI chatbots tend to answer the same question slightly differently each time, so a single poorer result is usually just a fluctuation. What is worrying is a decline that persists across several measurements or affects many questions at once.
How long should I wait before I consider a fall to be a trend?
Carry out a few more measurements using the same set of questions. If the result does not return to normal, whilst at the same time the number of citations for the page falls or a new competitor appears in the responses, then this is a trend, not a fluctuation.
Could updating the model reduce brand visibility?
Yes. The new version of the model may use different sources and organise responses differently. Such a drop usually affects a single chat and begins on the day of the update, so it is worth comparing the results before and after the update as separate periods.
Is the average result across all models sufficient to assess the decline?
No. A brand may lose ground in one chat and gain ground in another, and the average won’t reflect that. The decline needs to be assessed separately for each model.
Sources
- SparkToro and Gumshoe.ai, a study into the consistency of brand recommendations in AI (January 2026): https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/
- Aelo reports for Wapiennik „Węże” dated 27 July, 10 September and 14 October 2026 (published with the client’s consent)
- Aelo, GPT-6 in ChatGPT: what this means for brand visibility (October 2026): https://aeloapp.io/blog/gpt-6-w-chatgpt-widocznosc-marek/
Check on your brand
See which brands and models AI lists in your category.
250 AeloCoins to start with, no card required.
Author
Patryk Pietrzyk
Co-founder and CTO of Aelo, he is responsible for technology and app development. He runs the Blue Code software development studio. At Aelo, he decides how the system collects and processes responses from AI models.
More from the blog
Aelo
Check whether AI models mention your brand
Measurements taken on four models, using the same list of questions every month. 250 AeloCoins to start with, no card required.
From the same category
Newsletter
New research delivered straight to your inbox
We write when we have a new measurement or when something changes in the models.
Glossary
Report terminology explained in simple terms: mentions, citations, engagement with replies, and so on.
