A website change has not improved your AI visibility simply because the page was updated, indexed or looks better. To test whether AI answers changed, you need a recorded baseline, a fixed set of buyer questions, repeated measurements and a second measurement using the same method. You can then compare what changed — without claiming the website change caused it.

At TendorAI, we use a simple principle: Measure → Evidence → Find → Act → Remeasure → Compare. That means measuring what AI assistants say before making agreed changes, keeping a record of the work, then asking the same buyer questions again.

The important part is not merely measuring twice. It is keeping enough of the measurement unchanged that the two sets of answers can actually be compared.

Can you tell whether a website change improved your AI visibility?

Not from the website change alone.

If you rewrite a service page on Monday and ask ChatGPT about your business on Friday, a different answer does not prove the page worked. AI-generated answers can vary between repeated runs. Other websites change. Sources change. Competitors publish new information. AI systems themselves change.

A more defensible test is:

Record a baseline before making the change.

Freeze the buyer questions used in that baseline.

Record multiple answers to each question.

Keep the underlying answers and cited sources.

Record exactly what you changed and when.

Allow the changed information a reasonable opportunity to become available to AI systems.

Run the same measurement again.

Compare the two sets of recorded answers.

That tells you what changed between the two measurements. It does not, by itself, prove why it changed.

What is an AI visibility baseline?

An AI visibility baseline is a recorded starting point showing how AI assistants responded to a defined set of buyer questions at a particular time.

A useful baseline records more than whether the business appeared. It should preserve the questions, repeated answers, dates, assistants used, citations and underlying response text, so that a later measurement can be compared with the same starting point.

At TendorAI, three rules are particularly important.

1. Freeze the buyer questions

Once the baseline has been collected, keep the original questions unchanged for the comparison measurement.

Why? Because the question set is part of the measuring instrument. If Who provides X? becomes Which UK companies provide X?, you have changed more than the website. You have changed the question, and a different answer could reflect the different prompt rather than any change in the business's visibility.

New questions can still be added for exploratory research, but they should be reported separately rather than quietly added to the original comparison.

2. Repeat each question

One AI answer is an observation, not a pattern.

TendorAI's current measurement method uses five runs per question per assistant. Repeated runs matter because AI-generated answers can vary even when the question is identical.

If a business appears once in five answers before a website change and twice in five afterwards, that is an observed difference. It is not automatically evidence of a meaningful improvement. The underlying answers need to be examined as well.

3. Keep the recorded answers

Do not keep only a percentage or score. The underlying answers are the evidence. They allow you to check:

whether the business was actually mentioned;
how it was described;
whether it appeared in a list or a recommendation;
which pages or domains were cited;
which other businesses appeared;
whether the pattern was consistent across repeated runs.

A headline number can summarise a measurement. It cannot replace the evidence behind it.

Why every website change needs a change record

Suppose a business changes five service pages, publishes six articles, changes its company description and updates several directory profiles. Four weeks later its AI visibility changes. Which change mattered?

You don't know.

This is why TendorAI records agreed changes between measurement waves. A useful change record contains:

Record	What to capture
Page	The URL that changed
Change	Exactly what was added, removed or rewritten
Date	When the change became public
Reason	The evidence or finding behind the change
Measurement target	Which frozen buyer questions the change is intended to test

That last field is particularly useful. Instead of vaguely hoping that AI visibility improves, you can say: we identified a weakness on these buyer questions, made this specific change, and will examine those same questions in the next measurement. That is a testable process.

How long should you wait before measuring again?

There is no universal number of days that proves a website change has had enough time to influence AI-generated answers.

A changed page first needs to become available to the systems and sources involved in producing those answers. TendorAI therefore does not treat an arbitrary number of days as proof that a change has been incorporated. The decision to remeasure should consider what was changed, when it became public, and whether there has been a reasonable opportunity for the changed information to be discovered.

Even then, the next measurement tells you what the assistants said at that point in time — not exactly when or why they incorporated particular information.

Retrieved, cited, mentioned and recommended are not the same thing

This distinction is fundamental to TendorAI's methodology.

State	What it tells you	What it does not establish
Retrieved	A source was fetched or surfaced during the answer process, where that evidence is available	That its information affected the final answer
Cited	The final answer referenced a source	That the source caused the business to appear
Mentioned	The business appeared in the answer text	That the assistant recommended it
Recommended	The wording of the answer actually presents the business as a recommendation	Why it was selected

These states should not be collapsed into a single number.

For example, TendorAI has observed cases where one of its own pages was cited even though TendorAI itself was not named in the answer. That is useful evidence precisely because citation and mention are different events.

How should two AI visibility measurements be compared?

Compare the second measurement with the baseline question by question and assistant by assistant. At minimum, examine:

how often the business was mentioned;
the number of recorded answers behind that figure;
whether it was actually recommended;
which sources were cited;
whether the business's own pages were cited;
how the business was described;
which other businesses appeared in the answers;
consistency across repeated runs;
questions that improved;
questions that declined;
questions that did not move.

Always show the denominator. Four appearances from ten recorded answers is more informative than simply reporting 40% visibility.

And don't hide deterioration. If a business moved from 4/10 to 2/10 on one buyer question while improving elsewhere, both belong in the comparison. Otherwise measurement becomes marketing.

What can a before-and-after AI visibility comparison prove?

A before-and-after measurement can establish that the recorded answers differed between two measurement periods. It cannot automatically establish that a particular website change caused that difference.

That distinction matters. Between Wave 1 and Wave 2:

competitors may change their websites;
new sources may appear;
existing sources may change;
AI assistants may change;
the wider web may change.

An external measurement cannot hold all of those variables constant. The defensible statement is therefore: these were the questions we asked, these were the answers we recorded, these changes were made between the two measurements, and this is what was different when we measured again.

The evidence can support an association. It should not be turned into a causal claim it cannot support.

What if only one answer changes?

Record it. Don't ignore it, but don't exaggerate it either.

If TendorAI runs a question five times and a business appears once more in the second measurement, that is a real observed difference in the recorded answers. It is not enough on its own to establish a reliable improvement.

Look at the wider pattern:

Did the change repeat?
Did it occur across assistants?
Did relevant citations change?
Did the description of the business change?
Did related buyer questions move?
Did previously strong questions remain strong?

The evidence becomes more useful when those observations are considered together.

The TendorAI measurement and remeasurement process

TendorAI's managed process follows six stages: Measure → Evidence → Find → Act → Remeasure → Compare.

Measure — record how AI assistants answer the agreed buyer questions.
Evidence — preserve the answers, mentions and cited sources behind the measurement.
Find — identify where the business appears, where it does not, and what the evidence indicates should be investigated.
Act — make agreed changes to content, information or supporting evidence where appropriate.
Remeasure — ask the frozen buyer questions again using the same measurement method.
Compare — put the new answers beside the baseline and report what changed, what did not, and what requires further investigation.

That is the principle behind TendorAI AI Visibility Growth. It is not a promise that a website change will produce a recommendation. It is a way of making changes against recorded evidence and then measuring what happened afterwards. The measurement rules are set out in our methodology.

TendorAI is testing this process on itself

TendorAI collected its own AI visibility baseline in September 2026. That is a self-measurement, not independent validation.

We are using the findings to make controlled changes to our own website, and recording those changes before running the frozen questions again. When the second measurement has been completed, we intend to publish the comparison with its dates, denominators, methodology and limitations.

Until that measurement exists, there is no before-and-after result to publish. And that's the point. An invented success story is not evidence.

Frequently asked questions

How long after changing my website should I measure AI visibility again?

There is no universally reliable waiting period. The changed information needs a reasonable opportunity to become available to the systems involved before another measurement is useful. Record the publication date, and make the timing of the next measurement part of the methodology.

How many times should an AI visibility question be asked?

More than once. TendorAI currently uses five runs per question per assistant, so that a measurement is not based on a single answer. The same number of runs should be retained for a direct comparison.

Can I change the questions between measurements?

You can, but the changed question should not silently replace the original one in a before-and-after comparison. Keep the frozen question for the comparison and treat materially different questions as a new or exploratory series.

Does being mentioned mean an AI assistant recommends my business?

No. A mention means the business appeared in the answer text. Recommendation requires the wording and context of the answer to actually present the business as a recommendation.

Is being cited the same as being mentioned?

No. A source can be cited without the business being mentioned, and a business can be mentioned without its own website being cited. TendorAI records these separately.

Does improved AI visibility prove that the website work caused it?

No. It establishes what changed in the recorded answers between measurements. Other changes to AI systems, competitors and the wider web can occur during the same period.

Measure first. Change second. Measure again.

If you want to understand whether your business appears when potential customers ask AI assistants about services like yours, start with evidence rather than assumptions.

TendorAI AI Visibility Benchmark establishes the baseline. TendorAI AI Visibility Audit investigates what the recorded answers and sources indicate. TendorAI AI Visibility Growth turns agreed findings into controlled changes and measures again.

That gives you something much more useful than a promise: a record of what was said before, what you changed, and what was said afterwards.

How to tell whether website changes improved your AI visibility