How to Show Up in AI Overviews When You Don't Rank on Google
- Harold Bell
- Aug 9
- 14 min read
Key takeaways
|
Ask anyone how to show up in AI Overviews and you'll get a version of the same answer: rank first. Get to page one, structure the page properly, and the answer engines will pick you up. Ranking first, citation second.
We went looking for that relationship in our own data and couldn't find it.
In August 2026 we pulled all 165 posts on this site through the Wix Blog API and measured each one: word count, heading depth, question-form headings, answer boxes, internal links, schema types, publish date.
We joined that to every keyword we rank for in the Google top 100 from the Semrush US database, to every backlink we've earned, and to Microsoft's own record of every time Copilot used one of our pages to ground an answer. Then we asked a simple question. When somebody cites us, where does that page sit in Google?
Mostly nowhere.
With that said, we'll take you through our journey, and show you how to show up in AI overviews when you doin't frank in Google.
What Microsoft's own data shows
Over 90 days, Microsoft Copilot cited this site 602 times across 30 pages. 19 of those 30 pages don't rank in Google's top 100 for anything, and 80% of the citations came from pages outside Google's top 10. In the same period, Bing organic search has sent this site 4 clicks in its entire lifetime. |
Bing Webmaster Tools added an AI Performance report in February 2026. It's the only place a site owner can see first-party citation data from a major AI provider, and it counts how often Copilot pulled one of your pages in to ground an answer.
We opened ours for the first time after publishing the original version of this piece. Here's what it says.
Page | Copilot citations | Google position |
What is keyword gap analysis | 101 | 24 |
The complete guide to enterprise content marketing strategy | 78 | 5 |
Schema markup examples | 78 | 37 |
FAQ schema for AI search | 76 | 21 |
What to look for when hiring a SaaS marketing agency | 43 | 3 |
Why content marketing isn't producing ROI | 35 | Not ranking |
What is an AI marketing agency | 30 | Not ranking |
Best B2B content marketing examples | 20 | Not ranking |
Landing page best practices | 18 | Not ranking |
LLM optimization | 16 | Not ranking |
Our single most cited page ranks 24th on Google. The two pages we hold inside Google's top 10 account for 20% of citations between them. Everything else, 80% of the total, comes from pages sitting on page two or worse, or nowhere at all.
There's a second number in that report we didn't expect. Citation share, which is our citations on a query divided by everyone's, averages 21% across the queries Copilot grounded with our content, and reaches 43% on one. In Bing's organic results we are effectively invisible. In Copilot's grounding layer we're taking a fifth of the citations on our own topics.
What Copilot is actually asking
The report also exposes grounding queries, which are the internal search phrases Copilot generates to find sources. They're not what a user typed. They look like this:
keyword gap analysis definition explanation (87 citations)
FAQ schema improve citations AI search results (76)
AI marketing agency definition and overview (30)
internal team struggles before hiring B2B SaaS marketing agency (8)
reasons for inconsistent content performance quarter over quarter (3)
Those last two aren't keywords. They're problem statements, and no keyword tool would surface them. If you've been building content against a keyword list, this is a different map.
How many of our cited pages actually rank on Google
17 posts on this site have earned a deep editorial link from another publication. 11 of them (65%) don't rank for a single keyword in Google's top 100. Our overall library ranks at a rate of 27%, so cited pages rank slightly more often than average, but the majority of them don't rank at all. |
Here's a sample of who cited what, and where that page sits.
Who cited us | What they took | Page's Google position |
Our 40 to 80 word block data | Not ranking | |
A "3.2 times more AI citations" figure | Not ranking | |
Our entity authority framework | Not ranking | |
A numbered reference, anchor text "[14]" | Not ranking | |
Our FAQ formatting guidance | Not ranking | |
Our topical scoring model | 11 | |
A 40% figure | 36 | |
Our "broken schema scores zero" finding | 51 |
Look at the athenahq and snoika rows. Both are AI visibility tools. Both quoted a specific number off a page. Neither page ranks in Google's top 100 for anything.
The agilegrowthlabs link is the one that stopped us. The anchor text is "[14]". That's a numbered reference in a list of sources, the way you'd cite a paper. The page it points at, our entity authority piece, has never ranked for a keyword.
That's not a page being rewarded for ranking. That's a page being used because somebody needed the specific thing on it.
What we measured
Three datasets joined on the post slug, all pulled on 9 August 2026.
The structural pull came from the Wix Blog API and covers all 165 published posts with full rich content. For each post we counted words in text nodes, H2 and H3 headings, headings phrased as questions, colored answer boxes, internal and external links, images, and the JSON-LD types present.
The ranking data came from the Semrush US database using the organic positions report with SERP features enabled.
One thing worth stating plainly, because we got it wrong the first time. Semrush returns AI Overview citations as separate rows from organic rankings, and the position number on an AI Overview row is the slot inside the AI Overview, not the Google rank.
Read those rows as organic positions and you'll conclude that every AI citation comes from position one, which is the opposite of what the data says. We made that mistake, caught it, and it changed the finding completely.
The backlink data came from the Semrush backlinks report, filtered to links pointing at /post/ URLs so that directory listings, URL shorteners and our own homepage links don't inflate the count.
The limits of this dataset
One domain, 165 posts, B2B marketing, US English. Our Authority Score is 12. We have 52 referring domains and roughly 20 of them are shorteners and spam directories. We're not a representative enterprise site.
The samples are small. 17 editorially cited posts and three AI Overview citations. That's enough to say the relationship between ranking and citation isn't what people assume, and nowhere near enough to model what does drive it.
Semrush's SERP feature attribution is also a snapshot rather than a live query, and AI answers aren't deterministic. Querying the assistants directly would be a better measurement and it's the study we'd run next.
Why the citation rate collapses below position three
Citation is a cliff, not a curve. In positions one to three, three of our four AI Overview keywords are cited, a rate of 75%. From position four down to 100, across 179 AI Overview keywords, we're cited zero times. |
This is the finding.
Google position | AI Overview keywords | Cited | Citation rate |
1 to 3 | 4 | 3 | 75% |
4 to 10 | 6 | 0 | 0% |
11 to 20 | 13 | 0 | 0% |
21 to 30 | 30 | 0 | 0% |
31 to 50 | 58 | 0 | 0% |
51 to 100 | 78 | 0 | 0% |
We expected a decay curve. Something like 60% at the top, tapering through the teens, thinning out by page three. That's the mental model most of us carry, and it's the model that makes AEO work feel worthwhile at any ranking position: get a bit better, get cited a bit more.
We didn't find a curve. We found a step. Above position three, citation is likely. Below it, on our data, citation doesn't happen at all.
179 AI Overview keywords. Zero citations. Those pages are indexed. They're structured. Many of them are better written than the three that get cited. It doesn't matter.
The four positions that carry everything
We want to be careful here, because four observations is four observations.
What we can say is that the three cited posts share exactly one thing that the other 42 ranking posts don't: they hold position one. Not more words. Our SEO gap analysis post is 1,880 words, well below our library median. Not more schema. Not more internal links, where it actually has fewer than the uncited median.
If the position hypothesis is wrong, the alternative is that these three posts share some quality we didn't measure. That's possible. It's also the kind of claim that explains everything and predicts nothing, which is why we'd rather publish the position finding and let someone falsify it on a bigger sample.
Why ranking didn't predict citation
Across 186 AI Overview keywords where we hold an organic position, exactly one produced a citation, and it sat at position 11. Our seven positions inside the top 10 produced none. Microsoft's Copilot data points the same way at far greater scale: 80% of 602 citations came from pages outside Google's top 10. |
86% of the keywords we rank for sit on SERPs that generate an AI Overview, so the opportunity is nearly universal. What we can't find is the ranking relationship.
Our organic position | AI Overview keywords | Citations |
1 to 10 | 7 | 0 |
11 to 20 | 13 | 1 |
21 to 30 | 30 | 0 |
31 to 50 | 58 | 0 |
51 to 100 | 78 | 0 |
Then, outside that table entirely, two more citations on queries we don't rank for.
We want to be careful about how strong a claim that supports. Three citations is three citations, and you shouldn't build a strategy on it. But the direction is the opposite of the assumption. If citation followed ranking, our seven top-10 positions would be the likeliest place to find it. They're empty. The citations showed up at position 11 and at no position at all.
This is consistent with what we've argued elsewhere about AI visibility being a different discipline from search ranking, and it's consistent with the other half of the problem we wrote about in ranking but not being cited. Ranking well doesn't get you cited. It now looks like not ranking doesn't stop you either.
What it takes to show up in AI Overviews
Not depth. Our Copilot-cited pages carry a median of 4 H3 subsections against 5 for pages never cited, and 2 answer boxes against 3. The most cited page on this site is 1,734 words with no H3 subsections at all. Structure is not what separates them. |
We expected to find that cited pages were deeper, more heavily subsectioned, more thoroughly marked up. That's the AEO playbook and we've followed it. It isn't what the data says.
Feature | Copilot-cited (27 posts) | Never cited (138 posts) |
Median word count | 2,604 | 2,204 |
Median H2 count | 11 | 9 |
Median H3 count | 4 | 5 |
Median question-form H2s | 5 | 5 |
Median answer boxes | 2 | 3 |
Median internal links | 14 | 13 |
Carries FAQPage schema | 89% | 86% |
Cited pages have fewer H3s and fewer answer boxes. Question-form headings are identical. Internal links differ by one. The only feature with any gap is raw word count, and it's modest.
The individual cases are sharper than the medians. Our most cited page, with 101 citations, is 1,734 words with zero H3 subsections and two answer boxes. It's a below-median page by every structural measure we track. The page with 43 citations is 1,282 words, also with no subsections.
Meanwhile our content audit checklist runs 3,728 words with 18 subsections and 17 answer boxes, and has three citations. Our GEO optimization guide has 27 subsections and 10 answer boxes, and has never been cited at all.
If structure caused citation, that would be the wrong way round.
So what does explain it
We don't know, and we'd rather say that than invent a mechanism.
The honest observation is that the top cited pages are the plain, direct answer to a definitional or evaluative question somebody would actually ask. What is keyword gap analysis. What is an AI marketing agency. What to look for when hiring an agency. Does FAQ schema improve citations. The pages that lose are the ones covering territory where dozens of better-resourced sites have already published the definitive version.
That points at topic fit rather than page craft, which is a harder thing to action and a much harder thing to sell. It's also a hypothesis, not a finding. Testing it properly would mean comparing citation rates across topics while holding structure constant, on a library big enough to do it.
Why schema didn't separate them
89% of our Copilot-cited pages carry FAQPage JSON-LD. So do 86% of the 138 pages Copilot has never touched. The feature is too widespread across our library to explain anything, which is what happens when you apply a best practice to everything. |
We spent a long time making FAQPage schema and answer-first writing standard across the library. We succeeded. 142 of 165 posts carry it. And the result is that structured data now has no explanatory power on our own site, because there's almost nothing left to compare it against.
This isn't an argument against schema and we want to be precise about that. A page with no structure, no FAQ section, and a buried answer is unlikely to be pulled into a generated response. What our data can't support is the stronger claim, that adding schema markup causes citations. We added it everywhere and citations landed on 19 pages.
There's a real difference between "do this or you can't be cited" and "do this and you will be cited." A lot of AEO advice, some of it ours, blurs the two.
What this means if you're doing AEO work
Stop treating citation as a reward for ranking. On our data the two came apart completely. Write pages that contain specific, attributable claims, and make those claims easy to find under their own headings. |
Three things we're changing.
We're measuring citation separately from ranking
We'd been treating citation rate as a downstream metric, something that would improve as positions improved. It didn't. We're now tracking AI search visibility and organic position as two independent numbers, because on our own site they moved independently.
We're putting a specific number in everything worth citing
Every editorial link this site has earned came from somebody taking a specific claim off a page. A word-count range. A multiplier. A percentage. A named model. Not one of them came from a page being generally good on a topic. Our content length data and AEO audit checklist both got picked up because they contained a number somebody could quote.
We're rethinking what an unranked page is worth
120 of our 165 posts rank for nothing. Under the old model those are failures waiting on an SEO fix. Microsoft's data says otherwise. 19 of the 30 pages Copilot cites don't rank in Google at all, and one of them, a post on why content marketing isn't producing ROI, picked up 35 citations while ranking nowhere.
The reverse is also worth sitting with. 84% of our library has never been cited by Copilot once, and three pages carry 43% of everything. Not ranking doesn't mean invisible, and publishing more doesn't mean being cited more.
Bing indexing matters because it feeds retrieval, but coverage plainly isn't visibility. We have 86% of the library indexed in Bing, four lifetime clicks from Bing search, and 602 Copilot citations in a single quarter. Those three numbers describe the same site.
An unranked page is not necessarily an invisible page. That's uncomfortable for how most content programs are measured.
What would change our minds
We've already been wrong once in public here. The original version of this piece argued that depth explained citation, based on 19 cited pages. Microsoft's Copilot data covers 602 citations and says the opposite, so that claim is gone. The thesis it was supporting, that citation doesn't follow ranking, held up and got stronger. Sample size is why, and ours is still one domain.
A better instrument would help too. Semrush attribution isn't the same as querying assistants directly. The next study we'd run is a stratified query set across ChatGPT, Perplexity, Google AI Overviews and Copilot, multiple runs per query because the answers vary, recording whether we're cited as a linked source and which competitor is cited instead. That would also let us separate stable citations from ones that appear and vanish.
And our reading of the depth finding could be wrong. It's possible cited pages are simply older, or on topics that attract more writers, and that H3 count is riding along with something we didn't measure. If someone finds that citation tracks topic rather than structure, that's a better explanation than ours.
The full dataset, all 165 posts and every keyword position with structural features attached, is available on request. Take the numbers and check our work.
Frequently asked questions
Do you need to rank on Google to get cited in AI Overviews
Not on our data. Two of our three AI Overview citations came from queries where we don't rank in Google's top 100 at all, and the third ranked 11th. Meanwhile the seven positions we hold inside the top 10 produced no citations. Three observations is a small sample, but the direction runs against the assumption.
What percentage of cited pages rank in Google
On our site, 35%. 17 posts have earned a deep editorial link and 11 of them rank for nothing in the top 100. Our library overall ranks at 27%, so cited pages do slightly better than average while still mostly not ranking.
Does FAQPage schema help you get cited
We can't show that it does. 89% of our Copilot-cited pages carry FAQPage JSON-LD and so do 86% of the 138 pages Copilot has never cited. When a feature is on nearly every page it can't explain differences between pages. Our reading is that it's necessary rather than sufficient.
How do you show up in AI Overviews without ranking
On our data it happens without you doing anything to the page's structure. 19 of the 30 pages Microsoft Copilot cites don't rank in Google's top 100, and our cited pages carry slightly fewer subsections and answer boxes than our uncited ones. The pattern we can see is that cited pages tend to be the direct, plain answer to a question somebody would actually ask, rather than the most thoroughly optimized page on a topic.
What percentage of keywords trigger an AI Overview
Across the keywords we rank for, 86%, which is 186 of 217 organic positions in the Semrush US database as of 9 August 2026. That's specific to B2B marketing queries with informational intent and other niches will differ.
How do you measure AI Overview citation
We used the Semrush organic positions report with SERP features enabled, where feature 52 indicates an AI Overview. One important trap is that AI Overview citations come back as separate rows and the position number is the slot inside the AI Overview, not the Google rank. Reading it as a rank produces a completely wrong conclusion.
Is answer engine optimization worth doing
Yes, as a qualifying condition. Clean structure, answer-first writing and valid schema look like prerequisites. What our data pushes back on is the idea that doing them produces citations on its own, because we did them across the whole library and 19 pages have been cited.
Why does my content rank on Google but never get cited
Our data suggests ranking and citation are close to independent, so a high position won't produce a citation by itself. The pages of ours that get cited contain a specific, quotable claim under a findable heading. Pages that summarize a topic well without asserting anything specific give a writer nothing to take.
Can a low authority site get cited in AI Overviews
Yes, and more often than we assumed. At Authority Score 12 with 52 referring domains, Microsoft Copilot cited this site 602 times in 90 days across 30 pages, and our average citation share on those queries is 21%. Most of those pages don't rank in Google. Authority clearly isn't the only gate.
Does Bing indexing help with ChatGPT citations
Bing feeds ChatGPT's search retrieval so indexing there is worth doing, but it didn't produce visibility for us. 86% of our library is indexed in Bing with zero errors and the site has four lifetime clicks. Being indexed and being surfaced are separate outcomes.
How many posts on a site rank for nothing
We can only speak for ours, where it's 73%, or 120 of 165 posts. What surprised us is that some of those non-ranking posts are the ones earning citations, so we've stopped treating an unranked page as automatically wasted.
What would disprove this finding
A larger site showing citation rates that climb steadily with ranking position would break our reading. So would a direct multi-platform test finding our citations concentrated in pages that rank well. Both are worth running and we'd rather be corrected than confidently wrong.
How do you see your Copilot citation data
Bing Webmaster Tools added an AI Performance report in public preview in February 2026, under Search Performance. It shows total citations, which pages were cited, and the grounding queries Copilot generated to find them. It's free for any verified site and there's no click data in it, so a citation tells you your page was used to build an answer, not that anyone read it.
Comments