Recommendation Marketing is the discipline of making a business easier for buyers, and the AI systems assisting them, to understand, verify, and recommend. Recommendation Measurement is how we observe whether that standing changes without pretending every change proves why it happened.
The first requirement is a before picture taken before the work begins. That record is the baseline.
TL;DR
- A responsible AI visibility baseline is the documented first observation period on a Frozen Question Set, recorded under declared conditions before improvement work begins.
- It can show whether a business appears, how it is represented, what sources are cited, which alternatives appear, and how much the answers vary during the period.
- It cannot establish causation, measure human trust, predict future answers, or certify a business’s standing across every AI system and every buyer question.
- Once improvement work begins, the original before picture cannot be recreated honestly. You can begin measuring today, but you cannot manufacture yesterday’s baseline.
Why Recommendation Marketing Needs a Before Picture
Every claim of improvement contains a comparison. Visibility increased. Representation became more accurate. The business appeared in more relevant recommendations. Each statement compares a later state with an earlier one.
Without a documented earlier state, the comparison depends on memory, scattered screenshots, or a starting point selected after the results are known. None of those gives a buyer, client, board, or skeptical reader a reliable before picture.
This matters because AI-assisted research is already part of B2B buying. Gartner reported in May 2026 that 45% of 645 surveyed B2B buyers used generative AI during a recent purchase, primarily to gather information about vendors and products. The same study found that 69% preferred to validate AI-generated insights with a sales representative. Buyers are using AI, but they are not automatically treating its answers as settled fact.
Source: Gartner, “Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights,” May 20, 2026.
The baseline is therefore not a vanity score. It is a traceable starting record for the buyer questions that matter. The Frozen Question Set defines what gets asked. The baseline records what the answers were, under documented conditions, during a declared period, before evidence-improvement work begins.
Why the Answer Itself Matters
The buyer may never reach the underlying source. Pew Research Center analyzed 68,879 unique Google searches from U.S. adults’ browsing activity and found that 18% produced an AI summary. Question-shaped searches were much more likely to produce one: 60% of searches beginning with words such as who, what, when, or why generated an AI summary. When a summary appeared, users clicked one of its cited sources in only 1% of visits.
Source: Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results,” July 22, 2025.
That study is about Google searches, not all B2B buying or every answer engine. Its value here is narrower: when an AI-generated answer is present, many users do not click through to inspect the sources. That makes the business’s appearance and representation inside the answer worth observing directly.
What a Responsible Baseline Records
For every question in the frozen set, record the same five signal groups under the conditions named in the protocol:
Appearance. Does the business appear at all? If it does, is it recommended, listed as an alternative, mentioned incidentally, or used as a source?
Representation. What does the answer say about the business? Are its services, expertise, location, people, proof, and fit described accurately?
Citations. Which sources support the answer? Do those sources belong to the business, independent authorities, directories, publications, communities, or competitors?
Alternatives. Which other organizations occupy the answer, and in what role? These are competitors for consideration in this answer, even if they are not the companies on the internal competitor list.
Stability. Which observations repeat during the declared period, and which move? Variability is part of the record, not noise to delete because it complicates the story.
The protocol should also record the date, system, question wording, relevant settings or account conditions, and the response itself. The purpose is not to eliminate every variable. That is rarely possible in a live system. The purpose is to disclose enough of the conditions that later comparisons are intelligible and proportionate.
What It Can Tell You
A responsibly recorded baseline can show where the business stood on the defined buyer questions during the declared period. It can reveal recurring absence, inaccurate descriptions, thin or surprising citation patterns, unexpected alternatives, and answer variability.
Those observations help prioritize the work. Inaccurate representation may point first to entity clarity and corroboration. Repeated absence on a consequential question may justify a stronger answer and more visible evidence. Unexpected alternatives may change the competitive content the business needs to produce.
Those are diagnoses and priorities, not causal findings. A baseline can show what deserves investigation. It does not, by itself, prove why the observation occurred or which intervention will change it.
What It Cannot Tell You
- It cannot establish causation. The record shows what appeared under the stated conditions. It does not prove why a system produced the answer.
- It cannot measure human trust. Appearance, representation, citations, alternatives, and recommendation language are observable signals. Trust is a human outcome that requires other evidence.
- It cannot predict future answers. Models, retrieval systems, available sources, competitors, and buyer language continue to change.
- It cannot certify the whole market. The baseline applies to the frozen questions, systems, conditions, and period that were actually measured.
- It cannot become yesterday’s baseline after the work starts. If improvement work is already underway, begin a record now and label it accurately. It is a starting point from today, not a reconstructed picture of the past.
This is also why a single AI visibility score is not enough. A score may be useful as a summary if its questions, conditions, components, variability, and limits remain visible. When those disappear, the number may become easier to compare and harder to understand.
The Baseline Ends Before the Work Begins
The operational rule is simple: declare the baseline period, complete it, close the record, and then begin the improvement work.
There is no universal period that fits every question set or decision. A self-directed baseline should be long enough to include repeated observations under a schedule the business can actually sustain. A governed program should size the period and repetition to the decisions the measurement is expected to support. In both cases, the period and conditions are declared before the results are known.
Do not rewrite a service page halfway through the baseline because an answer was inaccurate. Do not add a new proof page because a competitor appeared. Record the finding. The correction becomes part of the work that follows.
What Right Angle Is Committing To
Right Angle is applying the same discipline to its own Recommendation Marketing work. Our Frozen Question Set is being completed. The baseline will be recorded before improvement work begins, and any results we publish will carry the conditions and limits needed to understand them.
That status matters. We are not claiming that the question set is already frozen, that a baseline exists, or that an improvement has been demonstrated. The commitment is to create the before picture before making the after claim.
Declare the Period Before You Change the Picture
If you already have a modest set of buyer questions, your next step is not to chase the first answer you dislike. It is to declare the period and the conditions under which you will observe those questions before changing the evidence environment.
You do not need to publish the results. Start by making the commitment visible inside your team: these are the questions, this is the observation period, this is what we will record, and no targeted improvement begins until the record closes.
If you declare a baseline period this week, tell me in the comments or directly. Not the findings. Just the commitment to create the before picture before you begin improving it.
Coming next: once a business can measure its standing more honestly, a harder question follows. What was the traditional agency model built to be accountable for—and where does that model now fall short?
Frequently Asked Questions
What is an AI visibility baseline in Recommendation Marketing?
It is the documented first observation period on a Frozen Question Set. It records appearance, representation, citations, alternatives, and stability under declared conditions before improvement work begins. Within Right Angle’s framework, that makes it the starting control for Recommendation Measurement.
How long should an AI visibility baseline last?
There is no universal duration. The period should be declared in advance, allow repeated observations, and be proportionate to the decision the record must support. The conditions and repetition matter more than adopting an unsupported standard number of days.
Can I improve my website while recording the baseline?
Routine business activity may continue, but targeted changes intended to influence the measured questions should wait until the baseline period closes. If material changes occur, document them and decide whether the record remains usable for the intended comparison.
Does a baseline tell me why my business is absent or misrepresented?
No. It records the observation. It may suggest what deserves investigation, but it does not prove causation. Multiple factors, including question wording, sources, system behavior, context, and normal variability, may contribute.
Can a baseline measure buyer trust?
Not directly. It can record observable signals such as recommendation language, citations, accuracy, and stability. Human trust, preference, and purchase decisions require separate evidence.
Is a baseline the same as an AI visibility score?
No. A score compresses multiple observations into a number. A responsible Recommendation Marketing baseline preserves the questions, conditions, underlying observations, variability, and limits. A score can summarize the record; it should not replace it.
Can I create a baseline after improvement work has already begun?
You can begin a measurement record today, but you cannot reconstruct an honest before picture from before the work. Label the new record as the starting point from today and avoid implying that it represents the earlier state.
B Randall Willis
Founder, Right Angle
Visibility + Engagement = Opportunity
