Marsh Appears in 96.0% of AI Answers. Here's Our Five-Step Plan to Improve Its AI Visibility.

Answer Capsule

In November 2026, Marsh appeared in 216 of 225 AI answers included in the Professional & Management Liability Insurance benchmark (96.00%). It received 42 recommendations (18.67%), with 25 Top 3 placements and 23 first-place recommendations. A mention means the company appeared in an answer; a recommendation means the answer affirmatively endorsed it.

Our plan compares the company with its competitors, checks conflicting AI claims against reliable sources, and identifies website pages, third-party sources and buyer questions to address. The historical results show how the figures changed across available months. Different monthly samples limit those comparisons, and conflicting AI claims require verification before corrections are proposed.

Current AI visibility

Questions This Section Answers

• How do Marsh mentions, recommendations and placements compare with its Professional & Management Liability Insurance peers?

Marsh recorded 216 mentions (96.00%) and 42 recommendations (18.67%) across 225 answers included in the benchmark. Top 3 placements numbered 25; first-place recommendations numbered 23. Framing among its mentions was 59 positive, 157 neutral and 0 negative.

Marsh appears in nearly every answer in the benchmark, so the immediate question is not whether the company is seen but how it is described when it is. The recommendation rate sits close to the two larger rivals, which suggests the gap is in the strength of endorsement rather than in reach.

The standout figure is first-place recommendations, which are unusually concentrated for Marsh relative to its shortlist placements. That pattern is worth protecting. The agency would build a scorecard that tracks reach, endorsement strength and placement separately, so the client can see which measure moves and which stays flat.

Framing is overwhelmingly neutral, with no negative mentions recorded. Neutral wording is not the same as endorsement, and the client should read it as room to sharpen the language around Marsh rather than as a problem to fix.

What changed since last month

Questions This Section Answers

• How did Marsh counts and coverage differ across the available monthly snapshots?

From October 2026 to November 2026, Marsh's mention count rose from 58 to 216, while its recommendation count rose from 24 to 42. The recommendation rate was 18.75% and 18.67% respectively.

The two monthly samples are not directly comparable: the November benchmark drew on a larger answer set and a different mix of AI services than October. The movement in mentions and recommendations should be read as a description of what was recorded, not as evidence of a trend or of any cause.

The recommendation rate held roughly steady across the two samples, while first-place and shortlist placements rose. Because the samples differ, the agency would treat the November figures as a fresh baseline rather than as an improvement to celebrate, and would avoid drawing conclusions until comparable samples are available.

Step 1. Measure

Questions This Section Answers

• What should the agency measure and verify first for Marsh?

AI answers contained 2 conflicting claims about Marsh: 0 critical, 0 high, 2 medium, 0 low. Of these, 0 were rated critical or high priority; the critical claims are included in that total. These ratings help prioritize fact-checking. They do not establish that a company or publisher made an error.

Two conflicting claims about Marsh's annual revenue appeared in the captured answers, both rated medium priority. Neither is critical, and neither has been verified. The useful first step is a scoped fact-review register that records each claim, its source, the date it refers to and the evidence needed to settle it.

A visibility comparison alongside that register would let the client see which claims actually travel with the company's name and which sit in low-traffic corners. That helps the client spend review effort where it matters rather than chasing every discrepancy. The agency would propose this as desk work with a named owner for each record.

Step 2. Make the website consistent

Questions This Section Answers

• Which recorded claim differences about Marsh should be verified before proposed page edits?

Both columns quote AI answers, not verified company facts. Resolve the entity, date and source before deciding whether a correction is warranted; different dates can explain an apparent conflict.

The revenue conflict is the clearest candidate for verification. Different reporting dates or different revenue definitions can produce an apparent conflict without either figure being wrong, so the entity, period and source need to be pinned down before anyone decides whether a correction is warranted.

The agency would prepare a page brief only after that check, and only if the client's own published figures support a clear statement. Where the evidence is ambiguous, the brief would recommend leaving the page alone and noting the decision. No correction request would be drafted from an AI claim alone.

Step 3. Align the external narrative

Questions This Section Answers

• Which cited third-party pages are relevant to fact review for Marsh?

These pages were selected because they relate to the company in the captured answers. Their current contents still need checking before any correction request. A page linked to a conflicting claim is not necessarily its cause, and the same answer may contribute to more than one row.

Several third-party pages appear repeatedly alongside Marsh in the captured answers, including broker rankings and revenue coverage. Those pages are worth reading in their current form before any request is considered, because a citation does not show that a page influenced an answer or that its contents are wrong.

The agency would review the shortlisted pages against primary sources, record what each currently says, and separate confirmed discrepancies from no-change decisions. Any publisher request would be evidence-backed and framed as a factual query, not an assertion of error. Responses and non-responses would be logged separately.

Step 4. Optimize priority company pages

Questions This Section Answers

• Which observed Marsh website pages should receive review briefs?

The company's website was cited 14 times across 12 distinct page addresses; the table shows 8 pages selected for review. Zeros in the answer columns mean no matching answers were recorded. A cited page is not automatically a page that accompanied a recommendation.

The company's own pages were cited only a handful of times, and mostly in answers that did not recommend Marsh. That is a modest footprint, and it points toward making a small number of priority pages clearer rather than rewriting the site.

Page briefs would use approved facts, answer the questions buyers actually ask, and state plainly what Marsh does in the relevant lines. The agency would check existing coverage first, since a new page is rarely the right answer when an existing one can be sharpened.

Step 5. Publish around buyer questions

Questions This Section Answers

• Which observed buyer questions support content briefs for Marsh?

These are recorded answer observations, not search volume, verified demand, or proof of missing website content.

The recorded questions are almost all ranking and comparison questions: who the top brokers are, who is best, and which broker suits a particular need. Marsh appears in nearly all of them, which means the content opportunity is about being the clearest answer, not about being discovered.

Briefs would address the comparison directly, with plain descriptions of Marsh's strengths and the criteria a buyer should weigh. The agency would confirm what the site already says before proposing anything new, and would treat these observations as answer patterns rather than as evidence of search demand.

Our 90-day action plan

Questions This Section Answers

• How should the agency sequence the proposed Marsh work and judge completion?

The first month establishes the baseline scorecard, the fact-review register and the page and source shortlist, each with an owner and a review status. The second month turns approved evidence into page briefs and justified publisher requests, with confirmed edits and no-change decisions recorded separately.

The third month produces buyer-question content briefs and a fresh scorecard. Completed work would be reported separately from measured answer outcomes, and the scorecard would be presented as a record of observed movement, with any link to specific work left open rather than assumed.

Methodology

Questions This Section Answers

• How are Marsh answer metrics and citation populations defined?

This report covers 2026-11. The exact dates when the original AI answers were collected have not been verified. The visibility rates use 225 answers that met the benchmark's inclusion rules. A company is recommended only when the answer both endorses it and uses positive wording. Top 3 and first-place counts are part of those recommendations, not extra recommendations.

An answer can mention or recommend several companies, so company counts cannot be added to find the number of distinct answers. The net sentiment score is positive minus negative mentions, divided by all company mentions; it describes answer wording, not customer satisfaction.

The wider citation analysis covers 700 captured answers and 3782 appearances of cited URLs. Citation appearances across all captured answers differ from the number of benchmark answers citing a page and the number that also recommend a company. The current contents of cited pages and disputed company claims still need independent verification.

Monthly sample coverage:

The answer samples differ between months. These comparisons show how the recorded results changed; they do not establish why they changed.

The monthly samples differ in size and in the AI services covered, so month-to-month comparisons describe what was recorded rather than why it changed. Recommendation requires both an endorsement and positive wording; placement counts sit inside recommendations rather than adding to them.

Answers can name several companies, so company counts do not sum to distinct answers. Cited pages and disputed claims still need independent verification before any correction is proposed.

Our first three priorities

First, settle the revenue question with primary sources and a clear date, then decide whether any page needs changing. Second, review the most frequently cited third-party pages in their current form and log confirmed discrepancies separately from no-change decisions.

Third, prepare buyer-question briefs for the ranking and comparison questions where Marsh already appears, checking existing pages before proposing new ones. Throughout, the agency would keep completed work and measured answer outcomes in separate columns so the client can judge progress without overreading it.