AI Citation Growth: 8-Week Data Analysis

Key Takeaways

Over eight weeks, written content drove citation share from 38 percent to 44 percent, a 79 percent growth rate that outpaced a category-wide 64 percent gain. Video content, tested for roughly two months, added zero AI citations and did not change the weekly growth slope. Share of citations, not raw citation count, is the only metric that holds up when category volume is rising across the board.

• Raw citation counts rose from 48 to 86 over eight weeks, but share growth (38 to 44 percent) is the figure worth defending.
• Written content grew citations 44 percent in the first four weeks, averaging 9.5 percent per week, a slope that held steady after video was added.
• Zero out of 14 AI-cited YouTube videos in tracked questions belonged to PushLeads, despite ranking number one on YouTube for one topic.
• Embedding video on 20 pages showed 18 flat, some pages gained impressions, and results trended slightly upward after 30 days, but the sample is too small to draw conclusions.
• Ranking first on YouTube did not guarantee an AI citation; being retrievable for the rephrased question the AI actually uses mattered more.
• Content topics must overlap with real buyer questions to enter the candidate pool that AI engines pull from.
• Video should not be purchased as an AI visibility shortcut without a documented before-and-after baseline.

What Did 8 Weeks of AI Citation Tracking Actually Show?

Eight weeks of AI citation tracking showed that written content consistently earned citations and grew share faster than the overall category, while video content produced no measurable citation gains during the same window. Starting June 23, the tracked account held 48 citations and a 38 percent share of citations within its category. By August 17, citations reached 86 and share climbed to 44 percent. That 79 percent growth in PushLeads citations outpaced the 64 percent growth recorded across the entire panel of competing pages in the same category.

The data collection method involved daily and weekly snapshots of all published content, with duplicates removed, to capture what Google’s AI Overview actually cited over time. This created a running series of data points rather than a single before-and-after comparison, which matters because AI citation pools shift as new content is indexed, category competitors publish more, and AI engines rephrase queries differently than the original search term typed by a user.

The lesson that stands out from the raw numbers is that share growth is more meaningful than total citation count. When an entire category grows, every player in it will see raw citations increase. The question worth asking is whether a specific page or domain is growing faster than the pool around it. In this case, the answer was yes for written content and no for video, at least within the two-month video testing window.

How Citation Share Differs From Raw Citation Count

Citation share measures a page or domain’s portion of all AI citations within a defined topic category, while raw count simply totals how many times any content on that domain appeared, making share the more accurate indicator of competitive position when category volume is rising.

When AI tools like Google’s AI Overview expand their citation pool, nearly every participant in a category benefits from rising raw numbers. A site that saw 48 citations one month and 86 the next might appear to be accelerating, but if the category as a whole grew at the same rate, that site is only keeping pace. In the eight-week data set, the category grew 64 percent while PushLeads citations grew 79 percent. The difference, roughly 15 percentage points of outperformance, is what pushed share from 38 to 44 percent.

That six-point share gain is the number worth reporting to a client or skeptical colleague because it holds up even after you account for tailwinds affecting everyone in the space.

Why do raw citation numbers keep rising even when content quality stays the same?

AI engines expand the surface area of topics they cover as their training and retrieval systems mature. More queries trigger AI Overviews than did six months ago, and each new query format creates additional citation opportunities. If a contractor category goes from 50 tracked questions to 80 tracked questions, every site in that category sees more chances to appear, regardless of whether their content improved. That is why share is the only defensible metric when presenting results over time.

What share percentage signals that content is outpacing the competition?

There is no universal threshold, but any share that is growing while the category also grows indicates the content is compounding advantage rather than simply riding a rising tide. In this data set, moving from 38 to 44 percent across eight weeks, while the category simultaneously expanded, represented meaningful separation. A site holding 20 percent share that climbs to 22 percent in a growing category is outperforming the average contributor by approximately 10 percent above the baseline rate.

The Weekly Growth Rate Before and After Adding Video

Before video was added to the content mix, citations grew from 48 to 69 over four weeks, a 44 percent gain that averages out to 9.5 percent per week. After video publishing began, the growth rate was 32.7 percent over the following weeks. The slope did not change in a way that could be attributed to video.

The same weekly trajectory that existed before video continued after it. That is the central finding of the video portion of the study, and it is an honest result because the data collection was not set up to flatter any particular content type. If the curve had bent upward after video was introduced, the data would have shown it. It did not.

One reason the slope stayed flat is almost certainly that the video content published during this period covered topics with little overlap with the buyer questions being tracked. Subjects like dashboards, software workflows, and internal process documentation do not compete for citations on questions like “how does a restoration company get insurance referrals” or “why isn’t my contractor business showing up on Google.” Relevant content must match the actual question before the AI engine even considers it a candidate for citation.

Does publishing more videos per week accelerate AI citation growth?

Volume alone does not move the needle if the topics do not match the questions buyers are actually typing. Publishing videos every week on topics that are adjacent to buyer intent but not directly answering buyer questions keeps that content outside the candidate pool the AI pulls from. The eight-week data showed zero of 14 AI-cited YouTube videos belonged to the tracked account, even though videos were being uploaded regularly throughout the period.

How long should a video content program run before drawing conclusions?

The transcript notes that 90 to 150 days of consistent content publishing appears to be the minimum window before content shows up broadly in AI citations. Two months, the duration of the video test in this data set, falls below that threshold. Calling video a failure at the eight-week mark is premature. What can be said accurately is that two months of video produced no measurable citation gains, and that the topics chosen during that period did not align with tracked buyer questions.

Why YouTube Ranking Number One Did Not Produce an AI Citation

A video that ranked number one on YouTube for a given query still received zero AI citations during the tracking period, because the AI engine appears to rephrase the original query before searching for sources, and that rephrased version did not retrieve the number one-ranked video as a candidate.

This is one of the more counterintuitive findings in the eight-week data. Conventional SEO logic assumes that ranking first on a platform guarantees visibility when that platform’s content is pulled into AI answers. The data here suggests the mechanism works differently. The AI generates its own version of the question, sometimes substantially different from the phrasing a user typed, and then retrieves content based on that internal rephrasing. A video optimized for the original query phrasing may not match the rephrased version the engine uses.

The distinction between “ranking” and “being retrievable” matters here. One tracked video ranked first on YouTube for its target phrase but was bypassed on August 17 when the AI used a differently worded version of the question. The video was findable when searched directly but was not retrievable under the AI’s own query construction. That gap is something most SEO and content strategies do not currently account for.

How does an AI engine decide which YouTube video to cite?

Based on available data from this tracking period, retrievability under the AI’s rephrased query version appears to matter more than traditional ranking position. A video that does not appear in the top 20 results when the AI’s internal phrasing is used is effectively invisible to the citation process, regardless of where it ranks under the user’s original phrasing. This means optimizing video content requires understanding how AI tools rephrase questions, not just how users type them.

Can a written page earn a citation even when the embedded video does not?

Yes. The data showed exactly this outcome. A written page about contractors using YouTube to generate calls earned a citation while the video embedded on that same page received no citation. The AI cited the text-based content on the page and separately cited a different creator’s video on the same question. The page was in the candidate pool; the embedded video was not. This suggests that written content and video content compete in separate retrieval processes, at least in part.

What the Video Embedding Test Showed at 30 Days

Embedding video on 20 pages and tracking impressions for 30 days produced mixed and inconclusive results: 18 pages showed no measurable change, some pages gained impressions, and the overall range shifted slightly from 7.4 to 9.3 on whatever impression scale was being tracked, but the 30-day window is too short to support a firm conclusion.

The test does provide a template that other sites can replicate with a longer runway. The methodology, embedding video on a defined set of pages, recording baseline impressions, and checking again at 30 and 90 days, is the right structure. The problem is that 30 days captures only the early phase of indexing and AI training cycles. Pages that saw no change at 30 days might show movement at 60 or 90 days as AI engines re-evaluate content signals.

For contractor websites and home-service businesses thinking about video as part of their content strategy, the practical takeaway is to run the test but extend the tracking window. Do not make investment decisions based on 30-day results. Build the baseline, add video to a defined set of pages, and check in at 30, 60, and 90 days before drawing conclusions.

Does embedding video improve organic search impressions for a page?

At 30 days across 20 pages, the data did not show a consistent improvement. Most pages were flat. Some showed small gains. The slight upward trend in the impression range, from 7.4 to 9.3, suggests a possible effect, but with 18 of 20 pages showing no change, the evidence does not support a strong claim. A 90-day follow-up on the same pages would provide considerably more reliable data.

Should a contractor add video to every page on their website?

Selectively embedding video on pages that already rank for buyer questions and receive meaningful impressions is a more defensible approach than blanket embedding across a site. The pages most likely to benefit are those answering specific service and location questions that buyers search before making a hiring decision. If a page is already being cited in AI Overviews for written content, adding a relevant video may strengthen its position over time, though the 30-day data here does not yet confirm that outcome.

Which Topics Actually Enter the AI Citation Candidate Pool

Content enters the AI citation candidate pool when it directly answers the specific questions buyers type, or the rephrased versions of those questions that AI engines generate internally. Topics that are only loosely related to buyer intent, such as software dashboards, internal workflows, or industry-adjacent subjects, do not compete for citation on the buyer questions being tracked.

Four of the five questions tracked in this study were not represented in YouTube’s top 20 results for those queries. If a video or page is not in the top 20 on the platform where AI retrieval begins, it cannot be cited. This makes topic selection a gating factor, not just a quality factor. A well-produced video on the wrong topic has zero probability of earning a citation on the questions being tracked.

For home-service contractors, this means the content calendar should start with a list of questions that buyers actually ask before hiring. Sources for that list include call recordings, form submission text, reviews that contain questions, and keyword research tools that surface informational queries. Understanding how many specific queries to target from the start is covered in more detail in this resource on how many keywords to focus on when doing SEO, which applies equally to AI citation targeting.

How do you find the buyer questions that AI engines are actually citing?

Track what questions appear in AI Overviews within your category over time. Record which sources are cited for each question, whether those sources are your own content, competitor content, or third-party content. After several weeks of consistent tracking, patterns emerge around which question formats trigger AI responses and which domains earn citations most frequently. That data set becomes the editorial guide for where to publish next.

What happens when content topics don’t overlap with tracked buyer questions?

The content is simply not evaluated for citation on those questions. It does not earn partial credit or benefit from proximity to relevant topics. The eight-week data illustrated this clearly: videos on dashboard and software topics received zero citations on questions about restoration companies, contractor marketing, or local SEO. There was no overlap in the candidate pool. Topic alignment is a binary condition for AI citation eligibility on any given question.

How This Data Applies to Disaster Restoration and Home-Service Contractors

Disaster restoration contractors and home-service businesses face the same citation dynamics shown in this data: written content answering specific buyer questions earns AI citations faster than video, and share growth within the category is the metric that reveals actual competitive position.

For restoration companies specifically, the buyer questions that trigger AI Overviews often involve insurance, emergency response timelines, and contractor selection criteria. These are information-dense topics that written content handles efficiently. A detailed page explaining how a restoration company works with insurance adjusters on a water damage claim can answer several related questions in a single piece of content, increasing the number of queries for which that page is retrievable.

The strategy of answering real buyer questions in written form is documented further in this disaster restoration marketing data guide, which covers the specific question categories that generate the most qualified leads for restoration contractors.

Video content is worth building, but the topic selection problem identified in this eight-week study applies directly to restoration and home-service contractors. Videos about operational topics, internal training, or software reviews will not earn citations on the buyer questions that drive service calls. Videos answering questions like “how long does water damage restoration take” or “what does a public adjuster do” have a far better chance of entering the candidate pool.

What citation share should a home-service contractor aim for within their category?

Any share above the average for the category represents a competitive advantage. In a category with 10 active content publishers, holding 15 to 20 percent share and growing it consistently is a reasonable target. The specific number matters less than the trend. A site moving from 10 to 14 percent share over eight weeks is outperforming the field by 40 percent above the average, which compounds over months into a durable citation advantage.

Building Attribution So You Know What Is Actually Being Cited

Attribution in AI citation tracking means confirming that the specific content being cited in AI Overviews is actually your own, not another creator’s content on the same topic that happens to appear alongside a mention of your site’s broader subject matter.

The eight-week data produced one example where a PushLeads page was cited alongside a video, but the cited video was not PushLeads content. Without an attribution check, that outcome might be logged as a video citation for PushLeads when it was actually a written-page citation that appeared near a third-party video. This kind of misattribution inflates the perceived performance of content formats that are not actually earning citations.

Building attribution means tracking three data points for every citation event: the question that triggered the AI response, the URL of the content that was cited, and who owns that URL. Guest posts and third-party features add a layer of complexity here, since a mention or link on an external domain might trigger a citation of that external page, not your own site. Understanding which placements actually drive citation share is essential for making accurate content investment decisions. The mechanics of how third-party placements affect citation visibility are related to the concept explored in this overview of high-quality guest posting websites and what makes an external placement worth pursuing.

How do you verify that your content is the source being cited, not a competitor’s?

Open the AI Overview response, locate the citation link, and confirm the domain and page URL are yours. If the citation links to a third-party article, a competitor’s blog, or a general reference site, your content was not cited regardless of topic proximity. Record that outcome separately. Over several weeks, you will have a clear picture of which of your pages are actually in the citation pool and which topics are being won by other publishers.

Does a citation of a page near your embedded video count as a video citation?

No. A page citation and a video citation are distinct events. If an AI Overview cites a written page and separately cites a video from a different creator in the same response, only the written page citation belongs to the page owner. The video citation belongs to whoever created the video. Logging both as wins for your content mix would overstate your video performance. Attribution requires checking the specific source URL for each cited element in the AI response.

Frequently Asked Questions

How many citations did the tracked account gain over 8 weeks?

The account started with 48 citations on June 23 and reached 86 citations by August 17, a gain of 38 citations over eight weeks. That represents a 79 percent growth rate for the tracked account, compared to a 64 percent growth rate for the broader category during the same period. The difference between those two growth rates is what drove citation share up from 38 to 44 percent.

Why is citation share more important than total citation count?

Total citation count rises when an entire category expands, regardless of individual content quality. Every publisher in a growing category will see more raw citations simply because AI engines are surfacing more responses. Share shows whether a specific domain is growing faster or slower than the surrounding pool. A share figure that increases while the category also grows confirms the content is compounding its position rather than just keeping pace with a favorable environment.

Did video content produce any AI citations during the 8-week period?

Zero out of 14 AI-cited YouTube videos in the tracked questions belonged to the account being studied. Video was added to the content mix approximately one month before the end of the eight-week window. The topics covered in those videos had minimal overlap with the buyer questions being tracked, which kept the videos outside the candidate pool the AI pulled from. The 90 to 150-day content maturation window had also not been reached for the video content.

What does it mean for an AI to rephrase a query before retrieving sources?

AI engines do not always use the exact phrasing a user typed when searching for sources to cite. They generate their own version of the question, sometimes with substantially different wording, and retrieve content based on that internal phrasing. A video or page optimized for the user’s original phrasing may not rank well under the AI’s rephrased version, which explains why a number one YouTube ranking still produced zero AI citations for one tracked video during this period.

How should a contractor decide which questions to target with written content?

Start with the questions that appear most frequently in call recordings, contact form submissions, and review text. Cross-reference those with keyword research tools that surface informational queries related to the service and geography. Then check which of those questions are already triggering AI Overviews and which domains are currently being cited for them. That intersection, high buyer intent, AI Overview activation, and weak current citation pool, identifies the best opportunities for written content investment.

What is the minimum test window before judging whether video helps AI citations?

The data from this study suggests 90 to 150 days of consistent video publishing is a more reliable evaluation window than the 60 days tested here. At 30 to 60 days, indexing and AI training cycles have not had enough time to register new content reliably. Decisions about whether to continue, expand, or discontinue video investment should be made with at least 90 days of data, and only if the videos are covering topics that match tracked buyer questions.

Can a page earn an AI citation even if its embedded video does not?

Yes. The data confirmed this directly. A page covering contractors using YouTube for lead generation earned a citation in an AI Overview. The video embedded on that same page was not cited. A different creator’s video on the same topic was cited separately. Written content and video content appear to be evaluated by separate retrieval processes in AI Overview generation, meaning page-level text quality, structure, and topic relevance can earn citations independently of any embedded video.

What should someone ask for if a vendor claims video improves AI visibility?

Request a documented baseline of citation share before video was added, a clear record of which questions were tracked, and a comparison of share at 30, 60, and 90 days after video publishing began. Confirm that the citations being counted are actually from the client’s own video content, not from third-party videos on related topics. If the vendor cannot produce a before-and-after citation share comparison with verified attribution, the claim of improvement cannot be evaluated.

Get Your Own AI Citation Audit With No Obligations Attached

The eight-week data set described here required consistent snapshot tracking, duplicate removal, share calculation, and manual citation verification across multiple content types. Most home-service contractors do not have the time or tools to build that tracking system while also running their business. An AI visibility audit covers the same ground: identifying which questions in your service category are currently generating AI Overviews, which of your published pages are actually being cited, and which questions you are missing from entirely. The audit produces a table showing your current citation share relative to competitors in the category, the specific URLs being cited for tracked questions, and a list of questions where competitor content is earning citations that yours is not. The report belongs to you whether or not any work follows from it. There is no pitch attached to receiving it. To request the audit, contact PushLeads and ask for the AI visibility report for your service category and geography.

Full Transcript

The complete spoken content of the video above, in text.