Why Cross-Industry AI Citation Benchmarks Can Mislead CMOs

Answer Capsule

A global AI citation benchmark can change the management interpretation of a market even when the underlying data has not changed.

Across the 282 markets assigned to the seven named macro sectors in the 283-market benchmark, changing the comparison from the overall study median to the relevant macro-sector median changed the above-or-below classification for:

• 21 markets, 7.45%, on persistence;
• 52 markets, 18.44%, on portability;
• and 25 markets, 8.87%, on the persistence-minus-portability gap.

The effect becomes larger when a more specific detailed-sector benchmark is used. Relative to the global median, the detailed-sector median changed classification for:

• 42 markets, 14.89%, on persistence;
• 67 markets, 23.76%, on portability;
• and 43 markets, 15.25%, on the gap.

Looking at persistence and portability together produces an even more practical result. 66 of 282 markets, 23.40%, changed descriptive quadrant when the benchmark moved from global to macro sector. Ninety-five markets, 33.69%, changed quadrant when the benchmark moved from global to detailed sector.

Across all three metrics, 116 of 282 markets, 41.13%, reversed at least one above-or-below classification between the global and detailed-sector benchmark. Twenty-nine markets reversed on at least two metrics, and seven reversed on all three.

These are not changes in performance. They are changes in interpretation caused solely by changing the comparison group.

The practical conclusion is therefore narrower than "never use global benchmarks":

Use the global benchmark to describe the broader AI citation environment. Use the industry benchmark to add commercial context. Use the actual market and platform data for decisions. Do not turn a cross-industry median into a client performance target.

This analysis extends the parent study, There Is No Universal GEO Strategy: How AI Citation Persistence and Portability Differ by Industry Across 283 Markets, by testing how often benchmark choice changes the classification of the same observed market.

Key Findings

One Number Can Produce Two Opposite Benchmark Labels

Questions This Section Answers

1. How often does an AI citation market switch from above benchmark to below benchmark when the comparison group changes?
2. Can a market be below the global AI citation median but above the median for its own industry?
3. Which AI citation metric is most sensitive to benchmark selection?

A benchmark is a comparison rule layered on top of an observed value. Change the comparison rule and the label can change while the value remains identical.

The overall 283-market medians were:

• 63.93% persistence;
• 22.70% portability;
• 39.68 percentage points for the persistence-minus-portability gap.

For each of the 282 named-sector markets, this analysis asked three times whether the market sat at or above the relevant median:

1. against the overall study median;
2. against its macro-sector median;
3. against its detailed-sector median.

The results show that portability is especially sensitive to benchmark choice.

The finding does not mean the industry benchmark is automatically "correct" and the global benchmark is "wrong." Each answers a different question.

The global median answers: How does this market compare with all qualifying commercial markets in the study?

The macro-sector median answers: How does it compare with markets inside the same broad commercial sector?

The detailed-sector median answers: How does it compare with a narrower set of markets that more closely resemble its commercial category?

The problem appears when those three questions are treated as if they were interchangeable.

Bankruptcy Lawyers flips on all three metrics

Bankruptcy Lawyers illustrates the issue cleanly.

Nothing about Bankruptcy Lawyers changes between those columns. Only the reference population changes.

A report that shows only the global benchmark would describe the market as below median on all three dimensions. A report that shows the Legal Services benchmark would describe the same observed market as above median on all three.

That is not a statistical curiosity. It is exactly the kind of framing difference that can alter an executive's interpretation of whether a result looks weak, typical, or strong.

Portability Is the Metric Most Likely to Be Misread by a Cross-Industry Benchmark

Questions This Section Answers

1. Which AI citation metric changes benchmark classification most often across industries?
2. Why can the same portability percentage look strong in one industry and weak in another?
3. Which industries show the largest global-versus-local portability benchmark reversals?

Portability produced the most classification reversals at both industry levels.

Relative to the global 22.70% portability median:

• 52 of 282 markets, 18.44%, changed above-or-below status against their macro-sector median;
• 67 of 282, 23.76%, changed status against their detailed-sector median.

The direction of those reversals matters.

Finance can look stronger globally than it does inside finance

Finance & Insurance had a 26.58% macro-sector portability median, nearly 3.9 points above the overall study median.

As a result, 24 of 78 finance markets, 30.77%, were at or above the global portability median but below the finance-sector median.

A global benchmark therefore makes a substantial group of finance markets look stronger relative to the full study than they look relative to their own industry.

The effect is even clearer inside Property, Auto & Commercial Insurance, where the detailed-sector portability median was 28.21%.

Seven of the 16 markets in that detailed sector, 43.75%, change portability classification relative to the global median.

Examples include:

A CMO could therefore receive a technically correct statement that a 25.8% portability result is "above the overall benchmark" while simultaneously receiving the equally correct statement that it is "below the insurance benchmark."

Both can be true.

Legal has the inverse problem

Legal & Compliance had a portability median of only 16.96%.

Nine of the 20 legal markets, 45.0%, were below the global 22.70% median but at or above the legal-sector median.

That means a global benchmark systematically places a large block of legal markets on the low side of the study even when those markets are at or above the center of their own observed industry distribution.

Bankruptcy Lawyers at 17.65%, Birth Injury Lawyers at 19.76%, Immigration Lawyers at 21.95%, Nursing Home Abuse Lawyers at 17.65%, and several other legal markets all fit this pattern.

Again, the point is not to make the industry comparison look more favorable. The point is to make the comparison explicit.

A Two-Metric Scorecard Changes for One-Third of Markets

Questions This Section Answers

1. How often does a market move to a different AI citation performance quadrant when the benchmark changes?
2. Can a CMO's high-persistence/high-portability label depend entirely on the comparison population?
3. Is a single cross-industry traffic-light score appropriate for persistence and portability?

Executives rarely interpret one metric in isolation. A more realistic scorecard might classify a market into one of four descriptive quadrants:

• high persistence / high portability;
• high persistence / low portability;
• low persistence / high portability;
• low persistence / low portability.

The underlying persistence and portability values do not change. Only the medians used to divide "high" from "low" change.

That produces substantial label movement.

So one in three markets receives a different two-axis quadrant label when the detailed-sector medians replace the overall study medians.

Several examples show why an executive dashboard should retain the benchmark label beside the score.

• AI Work Collaboration Platforms recorded 68.07% persistence and 29.03% portability. It is high/high against the global medians, but its 68.07% persistence sits below the 69.49% Business Software & Operations median, moving it to low persistence / high portability at the detailed-sector level.
• Air Purifiers recorded 69.49% persistence and 21.95% portability. It is high persistence / low portability globally, but 21.95% meets the Consumer Products, Retail & Digital Services portability median, moving it to high/high locally.
• Boat Insurance recorded 69.49% persistence and 25.79% portability. It is high/high globally but high persistence / low portability against the 28.21% Property, Auto & Commercial Insurance portability median.
• Dental Implants recorded 58.73% persistence and 7.53% portability. It is low/low globally, but its persistence is above the 57.48% Medical, Dental & Clinical Services median, moving it to high persistence / low portability locally.

None of these reclassifications proves better or worse marketing performance. They prove that a traffic-light label without a visible benchmark can be ambiguous.

Forty-One Percent of Markets Reverse on at Least One Metric Against the Detailed Sector

Questions This Section Answers

1. How many markets change at least one benchmark conclusion when a detailed industry comparison replaces the global median?
2. How often do multiple AI citation metrics reverse at the same time?
3. Can a global benchmark materially change the overall narrative of a market report?

Looking at persistence, portability, and the gap together gives a broader view of benchmark sensitivity.

When global medians are replaced by macro-sector medians:

• 79 of 282 markets, 28.01%, reverse on at least one metric;
• 15 markets, 5.32%, reverse on at least two;
• 4 markets, 1.42%, reverse on all three.

When global medians are replaced by detailed-sector medians:

• 116 of 282 markets, 41.13%, reverse on at least one metric;
• 29 markets, 10.28%, reverse on at least two;
• 7 markets, 2.48%, reverse on all three.

That is the strongest argument against converting the global median into a universal client score.

The global benchmark is still valuable. It allows broad cross-market context, gives researchers a common reference population, and helps identify structural differences across the study.

But for a client-facing conclusion such as "above benchmark portability," the report needs to answer a second question: above which benchmark?

A global comparison, a macro-industry comparison, and a detailed-industry comparison can all be legitimate. They should not be collapsed into one unlabeled status.

The Benchmark Can Also Change a Market's Apparent Percentile Position

Questions This Section Answers

1. How much can a market's apparent percentile position change when it is ranked within its own industry rather than across all industries?
2. Are benchmark reversals limited to markets sitting very close to the median?
3. Which AI citation metric shows the largest industry-relative percentile movement?

Binary above-or-below labels can exaggerate tiny differences around a median. To test whether the benchmark effect extends beyond simple threshold crossings, this analysis also calculated descriptive empirical percentile ranks.

Each market was ranked once against all 282 named-sector markets, then again inside its macro sector and detailed sector. Sector groups contain only 10 to 21 markets at the detailed level, so these percentiles should be treated as descriptive positions, not population estimates.

The median absolute detailed-sector shift was largest for portability at 12.04 percentile points. More than one-quarter of markets moved at least 20 percentile points on portability when ranked inside their detailed sector instead of across the named-sector population.

That means the benchmark effect is not confined to a few values sitting a fraction of a point around 22.70%.

The industry's source environment can materially alter where the market sits within the comparison distribution.

Global Benchmarks Can Be Too Harsh in One Industry and Too Generous in Another

Questions This Section Answers

1. Which industries are most likely to be understated by a global AI citation benchmark?
2. Which industries are most likely to look stronger globally than they do against their peers?
3. How large are the benchmark offsets between the overall study and specific industries?

The direction of benchmark bias is visible in the sector medians themselves.

Legal & Compliance

Compared with the overall study medians, Legal & Compliance was lower by:

• 17.41 points on persistence: 46.52% versus 63.93%;
• 5.74 points on portability: 16.96% versus 22.70%;
• 9.33 points on the gap: 30.35 versus 39.68.

A global-median scorecard therefore creates a relatively demanding reference point for legal markets. In this sample, 45% of legal markets changed persistence classification, 45% changed portability classification, and 40% changed gap classification when the legal-sector medians replaced the global medians.

That does not mean the legal benchmark should be lowered to make legal clients look better. It means a legal market should not be described as abnormal solely because it sits below a cross-industry median.

Property, Auto & Commercial Insurance

The detailed-sector portability median was 28.21%, 5.51 points above the global median.

Here the global comparison can be relatively generous. A market with portability around 25% to 27% is above the overall study median while remaining below the center of its own insurance peer group.

Business Software & Operations

The detailed-sector persistence median was 69.49%, 5.56 points above the global median. Its gap median was 46.63 points, 6.95 points above the global gap median.

That means a software market can clear the overall study benchmark without clearing the observed Business Software & Operations median.

For example, Accounting Software's 40.65-point gap is above the 39.68-point global median but below the 46.63-point detailed-sector median.

This is why a benchmark report should show the value, the comparison population, and the comparison median together.

A Benchmark Ladder Is Better Than a Universal Target

Questions This Section Answers

1. What benchmark hierarchy should a CMO use for AI citation persistence and portability?
2. When is a global AI citation benchmark useful, and when should it be replaced by a category-specific comparison?
3. How should agencies present industry benchmarks without turning them into promised performance targets?

A practical reporting structure is a benchmark ladder rather than a single threshold.

Level 1: The actual market

Start with the market being measured. Preserve its prompt scope, response support, platforms, dates, source portfolio cutoff, and recommendation observations.

This is the primary operational baseline because it is the environment the company is actually competing in.

Level 2: The detailed commercial sector

Use the narrowest defensible sector with enough observations. It can show whether the market is unusual relative to categories with a more similar source structure.

The 19 detailed sectors in this benchmark contain between 10 and 21 markets. They are useful descriptive contexts, not universal industry standards.

Level 3: The macro sector

Use the broad sector to show whether the detailed category is itself unusual. This is valuable for executive context and for organizations operating across several related categories.

Level 4: The overall benchmark

Use the 283-market benchmark for the broadest structural comparison. It is especially useful for explaining the study-wide persistence-portability pattern and for showing how far an industry's source environment sits from the commercial dataset as a whole.

The order matters.

A CMO should not start with 63.93% persistence as a target and ask why the company's market does not meet it. Persistence here is a market-level portfolio overlap statistic, not a brand KPI.

Similarly, 22.70% portability is not a target recommendation rate or an ideal cross-platform citation share. It describes the median market-level Jaccard overlap of eligible platform source portfolios under this study's methodology.

The companion article, How CMOs Should Set AI Search KPIs by Industry Instead of Using One Global Benchmark, develops the reporting model for brand-level KPIs. The parent study, There Is No Universal GEO Strategy, provides the full sector distributions.

What This Study Does Not Support

The benchmark-reversal analysis should not be used to claim that:

• one industry is easier to optimize for AI search;
• a lower industry median justifies a lower client target;
• a higher persistence or portability median predicts traffic, conversions, revenue, or recommendation share;
• industry membership causes a platform to cite particular sources;
• a brand below an industry median is underperforming;
• or moving above a benchmark would cause better commercial outcomes.

The metrics describe source-portfolio structure, not business outcomes.

A market can have high persistence because a stable group of domains recurs month to month. That does not tell us whether a particular brand is recommended.

A market can have relatively high portability because several platforms share source domains. That does not tell us whether the platforms give the same answers or rank the same companies.

A benchmark is therefore useful only after the unit of measurement is clear.

Methodology

The primary study uses the frozen July through September 2026 cohort of 283 qualifying tracked markets, 876,605 retained AI responses, and 8,718,629 citation occurrences resolved to registrable domains across ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Overviews, and Google AI Mode.

Portfolio ranks count each domain once per retained response and rank domains by response presence, with alphabetical tie-breaking. Eligible Top-100 comparison cells require at least 100 responses and at least 100 resolved domains. Thin or missing cells are excluded rather than assigned zero.

Persistence compares monthly domain portfolios. Portability compares platform domain portfolios. Both use Jaccard intersection divided by union and are summarized within each market from eligible pairwise comparisons.

The full benchmark medians are 63.93% persistence, 22.70% portability, and 39.68 percentage points for the within-market persistence-minus-portability gap.

This article uses the 282 markets assigned to the seven named macro sectors for benchmark-reversal calculations. The remaining qualifying market is retained as Unassigned in the overall 283-market benchmark and is not included in named-sector comparisons.

For each named-sector market, above-or-below classifications were calculated relative to the global median, the relevant macro-sector median, and the relevant detailed-sector median. "Above" includes equality with the median. A reversal means the same observed market is on opposite sides of two different benchmark medians.

The persistence/portability quadrant analysis combines those two binary classifications into four descriptive groups. A quadrant change means the observed persistence and portability values stayed fixed while the comparison medians changed.

The percentile analysis uses empirical ranks within the 282 named-sector markets, then within each macro and detailed sector. Detailed sectors contain 10 to 21 markets. Percentile movements are therefore descriptive sample positions, not estimates of a wider industry population.

Sector medians are unweighted medians of qualifying market values under the preserved taxonomy. Sector samples are unequal and non-randomized. The study is observational and does not establish causal industry effects.

Limitations

The benchmark is built from changing observed samples rather than a certified identical-prompt longitudinal panel. Of the retained responses, 104,071 used ingestion-date proxies. Only 246 markets had all three months and all six platforms eligible in the marginal portfolios.

Shared source ecosystems, retrospective taxonomy, candidate cross-job repetitions, changing platform behavior, and unequal market coverage limit population inference.

The benchmark-reversal counts are intentionally descriptive. A median split creates a simple management illustration, not a validated performance threshold. Small changes around a median can change a binary label, which is why the article also reports empirical percentile shifts.

Detailed-sector sample sizes are relatively small. They are suitable for descriptive context but should not be presented as precise population norms.

Citation authority is not recommendation authority. Domain presence does not verify ownership, page quality, editorial influence, traffic, conversions, or revenue.

Related Research

• There Is No Universal GEO Strategy: How AI Citation Persistence and Portability Differ by Industry Across 283 Markets
• How AI Citation Persistence and Portability Differ Across Finance, Software, Health, Legal, Home, and Consumer Markets
• How CMOs Should Set AI Search KPIs by Industry Instead of Using One Global Benchmark
• The Persistence-Portability Gap at Scale: 282 of 283 Markets Show Higher AI Citation Persistence Than Cross-Platform Portability Across 8.72 Million Citation Occurrences
• Citation Breadth vs Citation Depth: Which AI Authority Metric Should CMOs Care About?
• Can One GEO Strategy Work Across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews, and Google AI Mode?
• Which AI Citation Metrics Belong on a CMO Dashboard?

The earlier Persistence-Portability framework provides conceptual context. The 2026 AI Citation Authority Study covers an earlier high-stakes sample and is a separate population.