Ranking Isn't What Gets Insurance Agencies Cited by AI

The assumption that is actually costing you

Ask most agency owners why they have not touched their site's structure for AI search and you get some version of the same answer. Progressive spends more on marketing in a quarter than most independent agencies will spend in a career. State Farm owns page one for half the searches that matter. Why restructure a site for AI citation when a national carrier with a nine figure budget already owns the rankings those citations are supposedly built on.

That reasoning made sense as recently as last year, and it is exactly backwards now. It rests on a premise that used to be roughly true and has quietly stopped being true: that an AI Overview mostly just re-reads whatever already sits at the top of the organic results and repeats it back to you. If that premise held, outranking Progressive would in fact be a precondition for getting cited instead of them. It does not hold anymore, and the agencies still operating as if it does are leaving a specific, measurable opportunity on the table.

Here is the version of that reasoning we hear most often, almost word for word: "I write Medicare and ACA in one metro. Three national brands already own every top-10 slot for the phrases my clients search. Why would I spend a weekend restructuring my site for AI citation when I cannot even crack page one." It is a reasonable question if the underlying premise is true. It stops being reasonable the moment you learn that the biggest of those three national brands is, by an independent research firm's own count, capturing under three citations out of every hundred tracked in the entire industry. Owning page one for a phrase and owning the AI answer to that phrase turned out to be two different contests, and the data says plainly they are not won the same way.

What changed between rank and citation

For most of Google's history, one search produced one ranked list, and whatever appeared near the top of that list was, by definition, the material Google trusted most for that exact phrase. An AI Overview looked, for a while, like a compressed version of the same process: take the top few organic results, summarize them, cite them. Under that model, citation and rank were nearly the same thing wearing different clothes.

Google's own documentation describes something different now. It calls the mechanism query fan-out: the model generates "a set of concurrent, related queries" to pull in more information than the one visible search box ever asked for³. A shopper who types "how much does Medicare Part B cost" is not really asking one question to the system building the answer. Behind the scenes, the model may also be asking about IRMAA brackets, late enrollment penalties, and what changed for the current plan year, each of those a separate retrieval that can surface an entirely different set of pages than the visible search results ever showed.

That is the mechanical reason a page sitting on page four, or a page that would not crack the top 100 for the visible query at all, still gets pulled into the answer. It was not competing for the front page phrase. It was the best match for one of the sub-questions the model generated on its own, a question with a much smaller, much less contested field of candidates. Rank against the visible query stops being the whole game the moment the system is running several invisible queries behind it.

Run the same logic against an ACA question instead of Medicare and the pattern holds. A shopper types "how much is a silver plan with a subsidy for a family of four." The visible phrase is broad and contested; a dozen national brands and comparison sites have built pages chasing exactly that wording for a decade. But query fan-out is not limiting itself to that one phrase. It is plausibly also generating sub-questions about the 400 percent federal poverty level cliff, what counts as household income for the subsidy calculation, and what changes at the next open enrollment. A page that answers one of those narrower questions precisely, with the current plan year's numbers and a named source, is competing in a field of a few dozen candidates instead of a few thousand. That narrower field, not the broad phrase, is where an independent agency's page actually has a realistic shot.

The page built to rank vs the page built to be cited

Put the two approaches side by side and the difference in what each one is optimized for becomes obvious. Most agency sites, ours included in years past, were built for the first column. The data in this guide argues for spending new effort on the second.

What the data says now

Ahrefs put a number on how much this has shifted with a study of 863,000 keyword SERPs and 4 million AI Overview URLs, published March 2, 2026¹. Looking across every SERP block, only 37.9 percent of cited pages also ranked in the organic top 10 for the same query. Positions 11 through 100 accounted for another 31.2 percent, and 31.0 percent of citations came from pages ranking beyond position 100, meaning they were barely, if at all, visible in the classic search results a human would scroll through¹.

The comparison point makes the shift concrete rather than abstract. The same measurement, run in July 2025, found roughly 76 percent top-10 overlap¹. In under a year, the share of AI Overview citations coming from the pages you would expect, the ones sitting where a human eye would land first, dropped by half. Ahrefs' own read on the change: the system is "selecting far fewer pages straight from the original SERP," leaning instead on the sub-query results query fan-out generates¹.

Conductor ran the same question specifically inside insurance. Its report, last updated August 13, 2026, tracked more than 178,000 prompts and 3.6 million-plus insurance brand citations across seven AI engines between January and May 2026, with a dedicated AI Overviews measurement window from May 20 to June 20, 2026². The headline finding for anyone selling insurance online: a Google AI Overview appeared for 40.7 percent of insurance searches in the study, and 53.3 percent of those citations came from pages outside Google's organic top 10². Two different research teams, two different methodologies, one general and one insurance-specific, landed on the same direction: less than half of what gets cited was already sitting where classic rank would predict.

The two studies do not use identical methodology, and it would be dishonest to present them as one dataset. Ahrefs measured across all industries and all SERP block types, and defines "cited" as any URL appearing inside an AI Overview panel¹. Conductor measured insurance queries specifically, across seven separate AI engines rather than Google alone, with its AI Overviews figure isolated to a narrower June measurement window². Different questions, different scopes, and the raw percentages differ as a result, 37.9 percent top-10 overlap in one, 46.7 percent in the other. What makes them worth citing together is not that they agree on a single number. It is that neither one comes anywhere close to the near-total overlap you would expect if rank still decided the citation, and that is the only claim this guide is actually making.

Why no single brand owns this

The second half of Conductor's data is arguably the more useful half for an independent agency, because it answers the question that actually stops most owners from trying: if I cannot outspend the biggest names, is any of this winnable at all. Across 3.6 million-plus tracked insurance citations, the single largest named brand, Insureon, held 2.9 percent of citations. Progressive held 2.8 percent. The Hartford held 2.7 percent².

Read those three numbers together and the shape of the field changes. This is not a market with one or two names capturing the majority of the answer space the way one page usually captures position one in classic search. It is a field so fragmented that the biggest player in it, by a real research firm's own count, holds well under one citation in thirty. A market that fragmented is not a market where scale is the deciding variable. It is a market where structure and sourcing are, because those are the variables an independent agency can actually control on the same afternoon a national carrier's marketing department is still routing the request through legal review.

The practical implication for content strategy is specific, not just reassuring. It means the right unit of competition for an independent agency is the single page answering one question, not the brand as a whole. You are not trying to out-market Progressive's entire content operation. You are trying to have the better-sourced, better-structured page the next time a fan-out query asks about your exact plan type in your exact market, a contest that resets with every question and every metro rather than being locked in by whoever already has the biggest domain. That is a genuinely different, and genuinely more winnable, game than the one most agency owners think they are being asked to play.

What Google itself says about it

It is worth going to Google's own documentation rather than a secondhand summary, because the company has been unusually direct about this. Google's guide on optimizing for generative AI features, last updated July 10, 2026, states the eligibility bar in one sentence: "To be eligible to be shown in generative AI features on Google Search, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements"³. No mention of rank. No mention of domain age or budget. Indexed, and snippet eligible.

The same guide goes further and pushes back on the idea that a separate "AI SEO" discipline even exists: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO"³. Google is not describing a new game with new winners. It is describing the same technical fundamentals, crawlability, indexability, a clean and truthful page, applied to a retrieval system that now casts a much wider net than one ranked list.

That framing matters because it rules out two opposite mistakes. The first is assuming rank is everything, which the data in the last section already disproves. The second, just as common, is assuming AI citation is some exotic new specialty requiring its own separate tactics disconnected from ordinary technical hygiene. It is not. A page that is not indexed, blocked from a snippet by a stray max-snippet directive, or otherwise technically broken fails both the classic ranking test and the citation test at once, for the same underlying reason. We cover that specific eligibility mechanism, and how to check your own site for it, in a separate guide on why your insurance website doesn't show up in AI Overviews.

What actually predicts a citation

If rank is not the gate, something else is doing the selecting once a page clears Google's indexed-and-snippet-eligible bar. Across the sites we build and audit, the pages that get lifted and quoted share a short list of properties, none of which require a marketing budget to build.

None of that list is exotic, and none of it requires outranking anyone. It is the same list Google's own guidance points toward when it says generative AI search is "still SEO," just applied with the knowledge that the retrieval net is wider and the winning unit is a well-sourced answer to a narrow question rather than a page optimized to rank for a broad one.

A few of those six are worth slowing down on, because they are also where most agency sites quietly fail without anyone noticing. Answer-first structure sounds simple until you actually count sentences on your own service page and find three paragraphs of scene-setting before the number a reader came for. Schema that matches the page sounds like a technicality until you remember that FAQPage markup describing a question your visible copy never actually answers is not neutral, it is a mismatch a crawler can detect, and Google's own guidance singles out matching structured data to visible text as one of its few explicit rules for AI features. A "real" freshness date is the one agencies fake most casually, bumping a timestamp in a CMS without touching a word of the page, which is precisely the difference between a page that looks current and one that actually is.

How to check your own site this week

You do not need a research team or a paid tool subscription to see where your own pages stand against the six properties above. Most of this fits into an afternoon.

Pick five real questions your last five clients actually asked, in their own words. Not the broad phrase you would guess from a keyword tool. The actual sentence a client texted or said on a call. Those are the closest thing you have to the sub-questions a fan-out query would generate.

Search your own site for the direct answer to each one. If the honest answer is not on a page anywhere, or it is buried past two paragraphs of introduction, that page is not currently eligible to be lifted for that question no matter how the rest of your site performs.

Open the page's source and check whether the visible copy carries a source and a date for every number on it. A rate, a deadline, a coverage limit stated with no attribution is exactly the kind of claim a citation-seeking system has no reason to trust over a competitor who did name a source.

View your page's structured data with a free schema testing tool and read it against the visible page, side by side. Flag anything the markup claims that a visitor cannot actually see on the page. That mismatch is the one Google names explicitly, and it is a five-minute check most agencies have never run once.

Look at the "last updated" date you show a visitor, and ask honestly whether the content actually changed the last time that date moved. If the answer is no, the freshness signal on your site is decorative, not real, and a page reviewed for this article's plan-year figures should say so plainly.

Ask the assistant you use personally the same five questions, by name, and note whether your agency gets named as a source. That is not a scientific measurement, but it is the closest thing to ground truth you can check for free before deciding whether any of the fixes above are working.

None of these six steps requires new software or a developer, and the hardest one is usually the first, because most owners default to guessing a broad keyword phrase instead of writing down the actual sentence a real client used. Once you have the real questions, the rest of the check is mechanical.

What chasing the wrong metric costs

The practical cost of this mistake is not a wasted line item. It is a wasted year. An agency that spends its content and technical budget trying to out-rank State Farm or Progressive for a broad head-term phrase is competing on the one variable, budget and domain authority, where scale wins. Ahrefs' and Conductor's data both point at the same overlooked lane: 53 to 62 percent of the citation pool, by either study's measurement, is coming from pages that never won that broad-phrase fight at all¹².

Skipping that lane because it seems less prestigious than a page-one rank is how an agency spends real hours and real content budget chasing 38 to 47 percent of the available citation opportunity while ignoring the majority share. The fix is not spending more. It is redirecting the same effort toward the narrower, answer-first, sourced pages that the fan-out mechanism is actually built to surface, which happen to be exactly the pages an independent agency can produce as fast as, or faster than, a carrier's marketing department can clear one page through review.

There is a second, quieter cost worth naming. Search Console's Search Generative AI performance report, live on every property since June 2026, will show you AI Overview and AI Mode impressions on your own pages, but it still will not tell you which query triggered them or send you the click⁴. An agency with no sourced, structured content to show for a query where it should be citable is not just missing the citation. It is missing the ability to even diagnose the miss, because there is nothing in the report to point at except a page that was never built to be found in the first place.

It is worth being honest about what this does not mean. It does not mean every agency needs to rebuild its whole site this quarter. An agency with a handful of pages, a real answer to its clients' actual questions, and no glaring schema mismatch is probably fine running the six-step check above on its own and fixing what it finds without paying anyone. Where the math changes is scale and cadence: an agency running a dozen pages across two producers hits a ceiling on how much of this it can maintain by hand, especially the freshness and sourcing discipline, which is where an ongoing build or a managed cadence starts paying for itself instead of staying a nice-to-have.

How we build for this instead

Every site we build ships with the technical baseline Google's own guidance names as the actual gate: indexed, fast, crawlable, and snippet eligible by default, because we build on Astro and serve from the edge on Cloudflare rather than a page-builder platform that loads a runtime before your content ever paints. On top of that baseline, every page carries the schema stack this guide describes, Article with an organization author, FAQPage, and Dataset for anything citing outside data, matched to what is actually on the page rather than markup nobody wrote copy to support.

The free Audit scores exactly this: content depth and answer placement, question-and-answer formatting, E-E-A-T signals, schema markup, and the technical and agentic readiness this article's data says actually decides a citation, benchmarked against more than 56,000 audited agency websites⁵. It takes about a minute, costs nothing, and tells you specifically which of the six properties in the last section your own site is missing, rather than leaving you to guess.

Digital Foundation builds and maintains that structure on an ongoing basis, including a weekly or daily blog and location page cadence on the Pro and Scale tiers⁵, so the answer-first, sourced content this data rewards keeps publishing on a real schedule instead of arriving in one burst and going quiet. None of it depends on outspending a national carrier's marketing department. It depends on the page actually being built the way the data in this guide says the citation pool now selects.

What you get

Concretely, an agency that acts on this gets a site that clears Google's stated eligibility bar by default, a schema stack that actually matches its content instead of drifting from it, and a cadence of narrow, sourced, answer-first pages built for the sub-questions query fan-out actually asks, not just the broad phrase a carrier's budget already owns. It gets a real shot at a citation pool where the single biggest named competitor still holds under three percent, without needing a fraction of that competitor's marketing spend to compete for it.

It also gets something less tangible but just as real: a way to measure the work that is not "did I outrank the incumbent," a question an independent agency will keep losing indefinitely against a national ad budget, but "did I answer the specific question a real client asked, sourced and dated, on a page a system could actually find and trust." That is a question this size of agency can answer yes to on a normal week, and the data in this guide says it is the one that was actually being scored all along.

Questions agents ask