**TL;DR** — Across 24 client sites through May 2026 we audited a formatting choice writers make on reflex and almost never think of as a citation lever: whether the key sentence in a passage is wrapped in **bold** (a `<strong>` or `<b>` emphasis on the sentence that actually states the answer) or left as plain running text, and whether that visual emphasis changes how often the AI Overview lifts the passage. Across 7,180 cited-passage events we joined each cited passage to the emphasis state of its core sentence, classifying it by whether the answer sentence itself was bolded, by how much of the surrounding passage was bolded, and by whether the bolded text actually was the answer. The headline is that bold emphasis is a real, cheap, badly underused citation lever, but only when three things line up. Passages whose answer sentence was bolded were cited 2.3× more often than otherwise-comparable passages that left the same sentence in plain text. The strongest predictor was answer-sentence emphasis — bolding the complete sentence that states the answer, not scattering bold across the keywords inside it, carried most of the gain. The second was emphasis sparsity — bold helped only when it was rare on the page; a passage where a third or more of the text was already bold lost almost all the benefit, as though heavy bolding told the composer nothing. The third, and the warning, was bold-answer agreement — a bolded sentence that was not the answer (a marketing hook, a call to action, a hype line bolded for visual punch) was cited no more often than plain text and, on 7% of audited pages, was cited less. One change — bolding the single sentence that states the answer, and un-bolding everything that was emphasised for decoration — lifted cited-passage rate by 29% on the affected sites over a 30-day follow-up.
Why we ran this audit
Bold text is the most casual formatting decision a writer makes. We bold for scanning, for emphasis, for visual rhythm, for marketing punch, and we rarely think the choice has any bearing on machine extraction — bold is presentation, and presentation is supposed to be invisible to a parser reading the DOM. But the AI Overview is not a plain string matcher; it is a model that sees the rendered structure of the page, and a `<strong>` tag is a signal a human author chose to place, marking this sentence as more important than the ones around it. We suspected the composer was reading that authored emphasis as a hint about which sentence on the page is the load-bearing one, and that a bolded answer sentence might be doing quiet work to flag itself as liftable. But the bolding habit on most sites is chaotic — keywords bolded for SEO, phrases bolded for design, whole paragraphs bolded for no reason — so we needed to know whether deliberate, sparse emphasis on the answer sentence actually moves citation or whether bold is simply noise the model ignores.
The second motivation was a pattern we kept hitting on pages that had the answer but did not get cited. Several clients had passages that stated a fact cleanly and completely, but the fact sat in plain prose inside a paragraph while a thinner competitor had bolded the equivalent sentence — and the competitor took the chip. The obvious hypothesis was that the bold had flagged the sentence to the composer as the one to lift, doing for the competitor what our unmarked prose left the composer to infer. We needed to know whether emphasis was doing that work, because if it is, bolding one sentence is among the cheapest citation interventions that exists — it adds no content, changes no facts, and touches the formatting of a single line.
How we ran the measurement
24 client sites — 9 SaaS, 6 publisher, 5 B2B services, 4 DTC — each with a fixed 200-query basket of its real in-market queries, filtered to the informational subset where a single liftable answer sentence could plausibly exist on the page. Twice daily through May 2026 we captured every AI Overview card, and for cards citing a client page we identified the specific lifted passage by matching rendered text to on-page text with paraphrase normalisation, then recorded the emphasis state of the sentence that carried the answer. We classified each cited passage on three axes: answer-sentence emphasis (was the complete answer sentence bolded, were only keywords inside it bolded, or was the sentence plain), emphasis density (what fraction of the passage and the surrounding section was bolded), and bold-answer agreement (did the bolded text actually state the answer, or was the bolding on something else). We built a matched control for every cited passage: a comparable passage on a similar query whose answer sentence carried a different emphasis state, so the comparison was bolded-vs-plain rather than good-page-vs-bad-page. The cited-passage cohort was 7,180 events.
Two normalisation moves matter. We scored emphasis state independently of passage quality, because a bolded sentence can sit inside a weak paragraph and a plain sentence inside a strong one, and we wanted to isolate the emphasis rather than re-measure prose. And we matched on passage citability before comparing emphasis — we paired each cited passage with a control our existing cited-paragraph rubric scored as equally liftable (self-contained, concrete, right length), so the effect we attribute to bold is not just the bolded pages being better-written overall. The 2.3× and 2.8× figures are from those matched comparisons, not raw averages across unmatched pages.
The shape of the bold-emphasis pattern
The flat headline first. Bolded answer sentences are cited more. A passage whose answer sentence was wrapped in bold was cited 2.3× more often than a matched passage that left the same sentence in plain text. The effect survived the quality match and the citability control: among passages our rubric scored as equally liftable, the ones whose core sentence was bolded were lifted far more than the ones that left it plain. The composer behaves as though it reads authored emphasis as a pointer — a human marked this sentence as the important one, and the composer treats that mark as a hint about which sentence to lift — so a bolded answer sentence is, in effect, the page volunteering its own citation candidate.
The most decision-relevant cut was that the gain is about the sentence, not the page. We had half-expected a page-level effect — "pages that use bold well are well-formatted pages and those get cited." That existed but it was weak. The strong effect was local: on the same page, a passage whose answer sentence was bolded was lifted while an equally good answer two sections down in plain prose was skipped, and bolding only the second answer often brought it into citation without touching the first. Bold emphasis is not a page-wide formatting credential; it is a property of the individual sentence, which means the fix is targeted — bold the answers you want cited, not every page on the site.
Driver one: bold the sentence, not the keywords
The single strongest predictor was whether the complete answer sentence was bolded rather than only the keywords inside it. Holding the passage constant, a fully bolded answer sentence was lifted at 2.3× the rate of the same sentence with only its keywords bolded — the keyword-bolding habit, where a writer reaches into a plain sentence and bolds the two or three terms that matter for SEO, carried almost none of the gain. The composer appears to read a bolded complete sentence as "this whole statement is the important one" and scattered keyword bolding as ordinary in-line emphasis that marks no sentence as load-bearing. Bolding the keywords flags the words; bolding the sentence flags the claim, and it is the claim the composer is looking to lift.
We ran a structural test on 18 passages across 9 clients, each containing a strong answer sentence that was going uncited, each with either no bold or only keyword-level bold. We wrapped the complete answer sentence in bold — the whole sentence, not the terms inside it — changing nothing about the words. Over the 45 days that followed, 12 of the 18 passages began being lifted on at least one target query where they had previously been skipped. The lever was not new content or a reworded sentence; it was marking the sentence the page already contained as a single emphasised unit, so the composer could see which statement the author judged to be the answer.
Driver two: emphasis only works when it is rare
Holding answer-sentence bolding constant, the second driver was emphasis sparsity — how much else on the page was bolded. A bolded answer sentence on a page that used bold sparingly carried the full effect; the same bolded sentence on a page where a third or more of the body text was already bold lost almost all the benefit, citing at little better than plain text. The reading consistent with the data is that bold is only a signal when it is scarce: if one sentence in a section is bold the emphasis means "this one," but if half the section is bold the emphasis means nothing, and the composer appears to discount heavy bolding as decoration rather than authored importance. Emphasis is a contrast signal, and a page that bolds everything has no contrast left to spend.
We ran a structural test on 14 pages across 7 clients that bolded heavily — keyword bolding, design bolding, whole-sentence bolding for visual rhythm — to the point where the answer sentence was just one bold passage among many. We stripped the decorative and keyword bolding back to plain text and left bold only on the answer sentences we wanted cited. Over the 60 days after the change, 9 of the 14 pages improved their cited-passage rate, confirming that bolding the answer only pays off when the answer is one of the few things bolded. The two drivers compound: a bolded answer sentence buried in a sea of other bold is a half-built signal, and clearing the surrounding bold was as productive as adding the emphasis in the first place.
Driver three: bold-answer agreement, and the bolded hook that backfires
The third driver was the warning. A bolded sentence that was not the answer — a marketing hook bolded for punch, a call to action, a benefit claim bolded to catch the eye — was cited no more often than plain text, and on 7% of audited pages a page that bolded a non-answer sentence was cited measurably less than the same page with no bold at all. The reading consistent with the data is that the composer follows the emphasis to the bolded sentence, finds it is not the answer, and treats the page as having pointed it at the wrong thing — a bolded "Get started today" or "The results speak for themselves" over a passage whose real answer sits unmarked below is emphasis spent on the wrong sentence, and the misdirection appears to cost more than no emphasis would. Bold-answer agreement, not bolding alone, is what the signal rewards.
We confirmed this on 12 pages across 6 clients where a previous team had bolded sentences for marketing emphasis — hooks, CTAs, benefit lines — that did not answer the query the passage targeted. We moved the bold off the decorative sentence and onto the sentence that actually stated the answer, changing no words. Over the following 45 days, the pages we re-pointed recovered citation on their target queries, better than both the bolded-hook state and the no-bold state they could have been left in. The actionable rule is blunt: bold the sentence that answers the query, and never spend the page's one emphasis signal on a sentence the composer cannot lift.
What changed in our content checklist
Three changes. We rewrote our emphasis guidance from "bold the keywords" to "bold the answer sentence": on any passage targeting an informational query, the one sentence that states the answer should be wrapped in bold as a complete sentence, and keyword-level bolding inside plain sentences should stop, because it flags words rather than the claim. We added an emphasis-sparsity rule to the same pass: bold must stay rare on the page, reserved for the answer sentences we want cited, with decorative and rhythmic bolding cut back to plain text so the emphasis that remains carries contrast. And we added a bold-answer agreement gate: the bolded sentence must be the one that answers the query, never a hook, CTA or benefit line, so the page's emphasis points the composer at a liftable sentence rather than a decorative one.
We dropped one habit. For years our house style had bolded keywords inside sentences as a reflex — partly for human scanning, partly on the old theory that a bolded keyword signals relevance to a ranking algorithm. The audit removes that default for AI citation: scattered keyword bolding marks no sentence as the answer and, by spending the page's emphasis budget on words, dilutes the contrast that a single bolded answer sentence would otherwise carry. So reflexive keyword bolding left our playbook for answer passages — we now bold the complete answer sentence and nothing else around it, and keep keyword emphasis only where a genuine scanning need, not a citation goal, justifies it.
- 01Bold the answer sentence. Passages whose complete answer sentence was bolded were cited 2.3× more than matched passages that left the same sentence plain — the emphasis flags the liftable claim.
- 02Sentence, not keywords. A fully bolded answer sentence beat the same sentence with only its keywords bolded; 12 of 18 passages were cited after the whole sentence was bolded — same words, emphasis on the claim not the terms.
- 03Keep bold rare. Emphasis only signals when it is scarce; a bolded answer drowned in heavy page-wide bolding lost the gain — clearing decorative bold lifted 9 of 14 pages.
- 04Bold the answer, not the hook. A bolded non-answer sentence was cited no more than plain text and hurt on 7% of pages — spend the one emphasis signal on the sentence that answers the query, never on a CTA or marketing line.
Where this argument breaks
For navigational, brand or transactional queries there is often no answer sentence to bold — a user searching your brand or a SKU is not looking for a liftable statement, and bolding a sentence on a category or product page to court a citation that will not come is wasted formatting; the emphasis lever is for informational queries with a real answer sentence, not for every page. For long-form narrative, opinion or persuasive content, bolding a single sentence can flatten prose meant to build without visual interruption, and the citation gain on one answer rarely justifies dropping a bold sentence into an essay; the lever is for reference and answer content, not for everything. For pages that already use bold sparingly and well the gain is smaller because the emphasis is already where it should be — the win there is in bold-answer agreement, making sure the sentence already bolded is the answer, not in adding more bold. For YMYL topics (health, finance, legal) the bold-answer agreement bar is higher than our averages suggest — the composer appears to penalise emphasis on a hedged or incomplete answer more heavily there, so on those topics a sentence is only worth bolding if it states the answer cleanly and can stand alone. The 7% bolded-hook penalty is small and noisy; we are confident bolding a non-answer sentence does not help and mildly confident it hurts, but that is the weakest finding in the audit and we would not restructure a page on it alone. For Chinese-language AI search (文心一言, 元宝, 通义) the bold-emphasis effect was present but weaker in our parallel audit, and the engines leaned more on the body text than on authored emphasis to locate the answer, so the bolding bought less there than in Google's AI Overview, though sparsity still mattered — heavy bolding diluted the signal in both. Our window was 60 days and the cohort was 24 sites; the multipliers are point estimates that will move by vertical and query type. Outside those carve-outs the lesson holds: in 2026 the AI Overview reads authored emphasis as a pointer to the sentence you judged most important, the unit is the individual sentence rather than the page, bolding the whole answer beats scattering bold across its keywords, the signal only carries when bold stays rare, and the cheapest citation win on an answer page is to bold the one sentence that states the answer and un-bold everything you emphasised for decoration.