What Is an AI Divergence Candidate? Understanding Positive and Negative AI Recommendation Signals

Definition status: LLM Authority Index research classification used in the V0 AI Investor Signals framework.

Primary use: Flagging unusually strong positive or negative AI recommendation movement for follow-up research.

Current methodology version: V0

Answer Capsule

An AI Divergence Candidate is a company whose AI recommendation movement is strong enough under predefined measurement rules to warrant deeper commercial and financial follow-up.

In the current LLM Authority Index V0 framework, a Positive AI Divergence Candidate requires a recommendation-coverage increase of at least 5 percentage points, an exploratory 95% interval that remains above zero, and improvement across at least four AI platform families.

A Negative AI Divergence Candidate requires the reverse pattern: a recommendation-coverage decline of at least 5 percentage points, an exploratory interval that remains below zero, and deterioration across at least four platform families.

The word candidate is intentional. The label does not mean the company is financially improving or deteriorating. It does not mean a stock is undervalued or overvalued. It does not mean an investor should buy, hold, or sell a security.

It means the AI-side recommendation movement is unusual enough to be preserved and tested against later commercial outcomes.

Questions This Section Answers

• What qualifies a company as an AI Divergence Candidate?
• What is the difference between a Positive and Negative AI Divergence Candidate?
• How is an AI Divergence Candidate different from AI Visibility Market Divergence?

What Qualifies a Company as an AI Divergence Candidate?

The V0 classification combines three dimensions:

1. Magnitude
2. Directional uncertainty
3. Cross-platform breadth

A large point estimate alone is not enough.

A company can show a large change in recommendation coverage while still receiving a Mixed / neutral classification if the exploratory interval crosses zero or the movement is concentrated in too few platforms.

The current V0 rule is designed to avoid promoting every large-looking monthly move into a research signal.

Positive AI Divergence Candidate

A company qualifies when:

• recommendation coverage increases by at least 5 percentage points;
• the exploratory 95% interval remains above zero; and
• at least four AI platform families improve.

Negative AI Divergence Candidate

A company qualifies when:

• recommendation coverage decreases by at least 5 percentage points;
• the exploratory 95% interval remains below zero; and
• at least four AI platform families worsen.

Everything else remains Mixed / neutral under the current V0 watch-category rule.

The full methodology is documented in How We Measure AI Commercial Momentum.

Why Use the Word Candidate?

The classification is intentionally provisional.

A strong AI-side movement may later prove to be:

• commercially meaningful;
• temporary platform noise;
• a model or retrieval artifact;
• related to changing consumer demand;
• associated with product-level rather than parent-company behavior;
• already reflected in conventional data;
• or unrelated to later financial outcomes.

Calling the company a candidate preserves that uncertainty.

The label says:

This movement is unusual enough to study further.

It does not say:

This movement predicts the company’s financial future.

That distinction is central to the AI Investor Signals research program.

Why Magnitude Alone Is Not Enough

Suppose Company A shows a +9 percentage-point increase in recommendation coverage.

That may look important.

But imagine its exploratory interval ranges from -3 points to +21 points and only two of six platforms improved.

The observed point estimate is positive, but the measurement is noisy and platform concentration is high.

Under the V0 rule, that company would not become a Positive AI Divergence Candidate.

Now suppose Company B rises only +6 points, but:

• the interval remains above zero;
• five of six platform families improve; and
• sensitivity checks produce similar results.

Company B may provide a cleaner directional measurement even though its point estimate is smaller.

This is why the framework combines magnitude, interval direction, and platform breadth.

AI Divergence Candidate vs. AI Recommendation Momentum

AI Recommendation Momentum is the observed direction and magnitude of change in recommendation behavior.

AI Divergence Candidate is a classification applied when that movement meets predefined thresholds.

Every company with comparable observations can have recommendation momentum.

Only some companies become divergence candidates.

See What Is AI Recommendation Momentum?.

AI Divergence Candidate vs. AI Recommendation Coverage

Recommendation coverage is the level metric.

For example:

• July coverage: 28%
• September coverage: 36%

The recommendation-coverage change is +8 percentage points.

That change contributes to AI Recommendation Momentum.

The candidate classification then asks whether the movement also satisfies the interval and platform-breadth conditions.

See What Is AI Recommendation Coverage?.

AI Divergence Candidate vs. AI Visibility Market Divergence

These terms should not be confused.

An AI Divergence Candidate is an AI-side watch classification.

AI Visibility Market Divergence is a later-stage research framework that would compare a validated AI-derived commercial signal with contemporaneous analyst expectations, financial forecasts, valuation signals, or other market expectations.

The sequence matters:

AI measurement -> AI Recommendation Momentum -> AI Divergence Candidate classification -> prospective commercial validation -> comparison with investor expectations -> possible AI Visibility Market Divergence research

The V0 candidate label comes much earlier in the research process.

The full later-stage framework is described in AI Visibility Market Divergence.

Why Platform Breadth Is Part of the Rule

A company can show a strong aggregate change because one platform moved sharply.

That pattern may be real, but it is different from a change that appears across several independent AI systems.

The V0 candidate rule therefore requires directional movement across at least four platform families.

This is a breadth threshold, not proof of economic validity.

A four-platform move can still reverse.

A six-platform move can still be unrelated to financial outcomes.

The purpose is simply to avoid treating highly concentrated platform movement as equivalent to broad cross-platform movement.

For more on this distinction, see Recommendation Portability and Platform Concentration Risk.

Why the Exploratory Interval Is Part of the Rule

AI answers are variable.

Prompt families can also contain related observations that should not be treated as fully independent.

The investor framework therefore uses an exploratory uncertainty interval based on normalized prompt clusters.

The interval is not a causal confidence interval.

It is not a prediction interval for revenue or stock returns.

Its purpose is more modest:

Does the observed recommendation movement appear directionally stable enough within the current matched prompt structure to justify follow-up?

If the interval crosses zero, the V0 classification remains Inconclusive even when the point estimate is large.

Confidence Labels Are Separate From Candidate Labels

A company can be a Negative AI Divergence Candidate and still receive Medium rather than High AI-measurement confidence.

The watch category and confidence label answer different questions.

Watch category asks:

Is the movement directionally unusual enough to flag?

Confidence asks:

How strong is the underlying AI-side measurement given sample size, platform count, interval behavior, and cleaning sensitivity?

Keeping those labels separate prevents a company from receiving an implied financial score simply because it passed a directional screen.

Why Investors Might Care

Candidate labels create a disciplined follow-up queue.

Instead of reviewing hundreds of companies based on raw AI visibility, researchers can identify entities whose recommendation movement meets predeclared thresholds.

Those candidates can then be tested against later variables such as:

• branded search;
• website or app engagement;
• account openings;
• customer acquisition;
• loan originations;
• transaction activity;
• reported revenue;
• analyst estimate revisions;
• earnings surprises;
• and later market outcomes.

The candidate label is useful only if the next step is validation.

Without that step, it is simply a descriptive classification.

Positive Does Not Mean Bullish

A Positive AI Divergence Candidate does not mean:

• the company is undervalued;
• revenue will accelerate;
• earnings will beat expectations;
• the stock will outperform;
• analyst estimates will rise;
• or investors should buy the stock.

It means AI recommendation behavior strengthened enough to meet the current V0 watch rule.

That movement may later prove economically meaningful or irrelevant.

The research program is designed to preserve both outcomes.

Negative Does Not Mean Bearish

A Negative AI Divergence Candidate does not mean:

• the company is overvalued;
• revenue will decline;
• earnings will disappoint;
• the stock will underperform;
• customer demand is falling;
• or investors should sell the stock.

It means AI recommendation behavior weakened enough to meet the current V0 watch rule.

A company can post strong reported financial performance while its AI recommendation coverage declines.

That is exactly why time ordering and later validation matter.

What Would Strengthen a Candidate Signal Later?

Evidence would become more interesting if the movement:

• persists across additional months;
• remains directionally broad across platforms;
• survives alternative cleaning and capture specifications;
• appears in metrics close to the tracked brand or product;
• precedes changes in branded search or direct engagement;
• adds information beyond conventional operating data;
• and eventually contributes incremental explanatory or predictive value in walk-forward testing.

None of those steps should be assumed in advance.

What Would Weaken a Candidate Signal?

The candidate framework would be weakened if:

• large movements routinely reverse the following month;
• platform breadth does not improve reliability;
• AI recommendation changes show no relationship with later commercial behavior;
• any relationship disappears after conventional variables are included;
• entity mapping proves too weak for public-company inference;
• or results cannot be reproduced across additional companies and sectors.

A null result is part of the research design.

Methodology Notes

The current V0 watch rule uses:

• recommendation coverage change as the primary momentum metric;
• a minimum absolute change threshold of 5 percentage points;
• prompt-clustered exploratory uncertainty intervals;
• a minimum directional breadth of four AI platform families;
• separate confidence labels;
• separate presence and rank measurements;
• duplicate and capture sensitivity checks; and
• prospective freezing of the signal before later outcomes are observed.

Future versions may change these thresholds as the longitudinal dataset grows.

Any methodology change should be versioned.

Limitations

The thresholds are research rules, not natural laws

Five percentage points and four platform families are V0 design choices.

They may need revision after more history is available.

Candidate labels depend on the prompt universe

Different commercially relevant prompt sets can produce different recommendation behavior.

AI platforms change

Models, retrieval systems, indexes, and product surfaces evolve.

Public-parent mapping can be imperfect

A product or consumer brand may represent only part of a listed company.

Financial relevance is unvalidated

The candidate label has not been established as a predictor of revenue, earnings, analyst revisions, valuation, or stock returns.

Related LLM Authority Index Research

• AI Commercial Momentum Methodology
• What Is AI Recommendation Momentum?
• What Is AI Recommendation Coverage?
• Initial Findings From 25 Public Companies
• What Is an AI Investor Signal?
• AI Visibility Market Divergence
• Recommendation Portability and Platform Concentration Risk
• AI Investor Signal Tracker