The short answer
On short questions, sites with lots of backlinks still hold Google's top spots. On long questions, the kind buyers now type into AI tools, sites with under 1,000 referring domains took 50% of the top 3 (vs 31%) and 58% of AI Overview sources (vs 38%). The sources AI Overviews cited were fresher, and on the same results page more backlinks did not raise a page's odds of being cited. ChatGPT plays a different game: it links mostly to well-known vendor sites, and there backlinks do predict who gets the link.
Buyers now ask search and AI full questions, not keywords. We asked Google, AI Overviews and ChatGPT 400 B2B SaaS buyer questions, short and long: 3,266 page-1 results, 2,309 AI Overview sources, 1,592 ChatGPT links and 1,945 websites, collected by hand in the US market from 1 to 11 November 2026.
↓Download
Take the report and the data with you
Everything behind this study is free to use and quote, with a link back to this page.
Report · HTML
The full report
This page as a single file. Opens in any browser, charts included.
Download reportData · Excel · 0.8 MB
The full dataset
All 400 questions, 3,266 page-1 results, 2,309 AI Overview sources, 1,592 ChatGPT links and 1,945 websites, plus summary, intent and scorecard tabs.
Download ExcelData · CSV · 0.3 MB
Raw data as CSV
The 5 data tables as CSV files with a readme. Ready for Google Sheets, Python or R.
Or one at a time: Questions · Google page 1 · AI Overview sources · ChatGPT links · Websites · Readme
Download all (zip)01Abstract & key findings
What we found, in one paragraph
We wrote 200 real B2B SaaS buyer questions, then wrote a long version of each (median 22 words vs 6) that adds the buyer's situation. We ran all 400 on Google, recorded every AI Overview source and asked ChatGPT each one. On the long versions, small sites took 50% of top-3 spots (up from 31% on the same topics) and 58% of AI Overview sources (up from 38%). 72% of AI Overview sources showed a date (vs 25%), with a median age of 130 days (vs 178). Across all 400 questions, 10x more referring domains gave a page 0.80x the odds of being cited in the AI Overview, so no lift. For ChatGPT the same test gave 3.64x: it favours established vendor brands.
Key findings
- Long questions open the door to small sites. Sites under 1,000 referring domains held 31% of top-3 spots on short questions and 50% on the long versions of the same topics. The share rose on 105 topics and fell on 24.
- AI Overviews follow. 58% of sources came from small sites on long questions, vs 38% on short ones. The median cited site had 557 referring domains, down from 4,213.
- Backlinks do not buy AI Overview citations. Compared with other pages on the same results page, 10x more referring domains gave 0.80x the odds of being cited across 400 questions (0.85 short, 0.79 long).
- Backlinks sort Google less on long questions. The correlation between backlinks and rank fell from -0.23 to -0.14. The site with the most backlinks ranked #1 in 20.5% of short searches and 14.0% of long ones.
- Fresh pages lead on long questions. 72% of AI Overview sources showed a date (vs 25%). Of those, 41% were under 3 months old (vs 32%) and only 14% were over a year old (vs 26%).
- The newest pages sit at the top of the answer. The first 3 sources had a median age of 117 days; later sources, 149 days.
- A new year in the title is not freshness. Pages with a year in the title were cited 33% of the time vs 56% without (0.65x the odds). 28% of cited "2026" pages carried a date over 6 months old.
- The page decides, for both engines. Product and landing pages: 3.37x the odds of an AI Overview citation and 4.79x for a ChatGPT link. A number in the title: 0.37x and 0.57x.
- ChatGPT is the exception on backlinks. Company sites under 1,000 referring domains were linked 4% of the time they ranked on page 1, vs 28% for sites with 1,000 to 10,000. 98% of its links went to company sites, mostly homepages (50%) and pricing pages (23%).
- Each version of a question gets its own results. Only 24% of page-1 sites and 21% of AI Overview sites were shared between the short and long version of the same topic.
02Methodology & sample
How the data was collected
Everything was collected by hand in a regular browser between 1 and 11 November 2026, for the US market (US English Google results, ChatGPT asked for US context). No paid SEO tools or APIs were used.
| Part | What we did | Size |
|---|---|---|
| Short questions | B2B SaaS buyer questions written the way a buyer types into Google: 100 "best X for Y", 30 "X alternatives", 40 how-to and what-is, 30 pricing and "X vs Y". Median 6 words. | 200 |
| Long questions | A long version of each short question, written the way people now ask AI tools: the same topic plus team size, budget, tools in use or the job to be done. Median 22 words. Each long question is paired with its short version. | 200 |
| Google page 1 | Every organic result on the first page, one row per website, with its position, title and URL section. | 3,266 |
| AI Overviews | Every source listed in the AI Overview. Usable on 394 of 400 questions (a few short questions got an answer about a different meaning of the words, and 2 long questions showed no AI Overview). | 2,309 sources |
| ChatGPT | Each question asked in a fresh, signed-out chat with web search on. Every outside link recorded. Free-tier message limits cut some answers. | 372 answers, 1,592 links |
| Backlinks | Referring domains for every website from a free public link index (the Open PageRank top 10 million domains list, built from Common Crawl). | 1,945 sites, 95% found |
| Labels and dates | Each result tagged by site type, page type, title traits and URL section. Dates are the ones Google shows on AI Overview source cards; YouTube upload dates are not counted. | 20+ traits |
To separate real effects from topic effects, the main tests compare pages only against other pages on the same results page (a conditional logistic regression). A "best CRM" page is never compared with a "SOC 2 pricing" page. The short vs long comparisons pair each topic with itself.
03Definitions
Terms used in this report
- Referring domains (RD)
- The number of different websites that link to a site. This is our backlink measure. It counts the whole site, not one page.
- Small site
- A site with under 1,000 referring domains.
- Short and long question
- The same buyer topic asked two ways: a short search (about 6 words) and a full question with the buyer's situation (about 22 words).
- Page 1
- The organic results on Google's first page. Ads are left out.
- AI Overview
- Google's AI answer at the top of the results. A site is cited when it is listed among the answer's sources.
- Linked by ChatGPT
- ChatGPT's answer includes a link to the site.
- Odds ratio
- How a trait changes the odds of being cited or linked, compared with other pages on the same results page. 1 means no effect, 2 means double the odds, 0.5 means half.
- Not sig.
- The result could be chance (p-value above 0.05).
- Breadth
- How many of our questions a site ranks on page 1 for. A rough measure of how widely a site covers the topic.
- Community sites
- Reddit, YouTube, Quora, LinkedIn posts, Medium and similar sites where users post the content.
04Long questions
Longer questions, smaller winners
People now ask AI tools full questions instead of typing a few keywords. We took each of our 200 topics and asked it both ways. On the long version, small sites took 50% of Google's top 3, up from 31%, and 58% of AI Overview sources, up from 38%.
Long question (median 22 words)Short question (median 6 words)
200 paired topics. Small sites in top 3: p < 0.001; small AI Overview sources: p < 0.001 (paired Wilcoxon test).
1,574 and 1,692 page-1 results; 1,137 and 1,172 AI Overview sources.
- It is a real shift, not noise. The share of small sites in the top 3 rose on 105 topics and fell on only 24. For AI Overview sources it rose on 140 topics and fell on 33.
- Long questions get mostly different results. Only 24% of page-1 sites and 21% of AI Overview sites were the same for the short and the long version. Ranking for the short keyword does not carry over.
- It is the long form that matters. Within the long questions, adding more details (team size, budget, tools, a constraint) did not raise the small-site share further. Section 05 has the numbers.
- Community sites shrink. YouTube appeared in 40% of long-question AI Overviews vs 78% on short ones. Reddit was on page 1 for 123 long questions vs 155 short ones. Company pages written for the specific situation take that space.
Small sites at the top on long questions
| Long question (topic) | Site | Position | Referring domains | Cited by AI Overview |
|---|---|---|---|---|
| Sales playbook for a 10-rep B2B SaaS team | ricavi.ai | 2 | 3 | Yes |
| SEO tool for a small B2B content team on a modest budget | rankup.so | 1 | 13 | Yes |
| HubSpot or Salesforce for a 40-person startup with no admin | delverise.com | 2 | 12 | Yes |
| How to write a B2B case study sales teams can use | edifycontent.com | 1 | 17 | Yes |
| Webflow or WordPress for a marketing team without developers | digihotshot.com | 2 | 17 | Yes |
| Digital adoption platform with tours and checklists | usertour.io | 1 | 21 | Yes |
| Hotjar alternatives with free or low-cost plans | thefrontkit.com | 2 | 18 | Yes |
Our reading: a long question names a team size, budget or use case, and a page written for that exact situation matches it better. Those pages are often on smaller, newer sites. The data shows the shift; it cannot prove the reason.
05By intent
The shift is biggest on learning questions and smallest on pricing
We sorted all 200 topics by what the buyer wants: to learn (definitions, benchmarks, how-to), to build a shortlist ("best X", alternatives) or to decide (X vs Y, pricing). Small sites gained 30 points of the top 3 on learning questions, 20 on shortlist questions and only 8 on decision questions.
Long questionShort question
Paired tests by stage: learn 24% to 54% (p < 0.001), shortlist 33% to 53% (p < 0.001), decide 29% to 37% (p = 0.23, could be chance).
| Intent | Small sites, top 3 | Small sites, AI sources | Median age, long |
|---|---|---|---|
| Definition or benchmark | 21% → 48% | 27% → 57% | 197 days |
| How-to | 26% → 58% | 35% → 62% | 249 days |
| Best tool for a need | 35% → 53% | 38% → 56% | 109 days |
| X alternatives | 29% → 53% | 38% → 49% | 132 days |
| X vs Y | 25% → 37% | 47% → 55% | 132 days |
| X pricing | 36% → 38% | 47% → 72% | 60 days |
What gets cited for each intent (long questions)
| Intent | AI Overviews cite most often | ChatGPT links per answer |
|---|---|---|
| Definition or benchmark | guides and articles (70% cited) | 2.1 |
| How-to | "best X" list posts (70% cited) | 3.2 |
| Best tool for a need | product pages (71% cited) | 4.3 |
| X alternatives | videos (75%) and Reddit threads; comparison pages 34% | 5.4 |
| X vs Y | comparison pages (65% cited) | 2.4 |
| X pricing | the vendor's own pricing page (cited on every question) | 1.0 |
- Decision questions stay with the vendors. On "X vs Y" questions, the named brands' own sites were cited in 65% of long-question AI Overviews. On pricing questions the vendor's site was cited every time (100%). Small sites still filled 72% of the other pricing sources.
- On "X alternatives", X is left out. The brand named in the question was cited in 7% of short and 0% of long AI Overviews. The space goes to rivals, comparison pages, videos and Reddit.
- Freshness follows the buying stage. Median age of dated sources on long questions: 98 days for decision questions, 114 for shortlist questions, 210 for learning questions.
- Backlinks did not help AI Overview citations at any stage. On the same results page, with page traits counted, 10x more referring domains gave 0.77x the odds on learning questions, 0.85x on shortlist questions and 0.64x on decision questions. For ChatGPT the same figures were 4.65, 2.99 and 1.51*.
- It is the long form, not the number of details. Within long questions, adding more details (team size, budget, tools used, a constraint) did not raise the small-site share further: 53% of top-3 spots with no extra details, 52% with two, 41% with three or more.
- ChatGPT links the most on shortlist questions. 4.5 links per long answer, vs 1.8 on decision questions and 2.8 on learning questions, where 25% of answers linked nothing.
How we sorted: by the wording of each question (what is, how to, best, alternatives, vs, pricing), not by our original ID ranges. * ChatGPT figure with position counted only; too few questions to count page traits.
06Backlinks & Google rank
Are backlinks still important? A little on short questions, less on long ones
| Measure | Short questions | Long questions |
|---|---|---|
| Correlation of backlinks with position (−1 = perfect) | -0.23 | -0.14 |
| Site with the most backlinks ranked #1 | 20.5% | 14.0% |
| Odds of a top-3 spot per 10x more referring domains | 1.51 | 1.20 |
| Median referring domains in the top 3 | 4,932 | 935 |
| Top-3 spots held by small sites | 30.8% | 50.5% |
- On short questions the top-3 lead is mostly Reddit, YouTube and big brands. With community sites removed, the short-question top-3 median drops from 4,932 to 1,476 referring domains.
- On long questions backlinks barely sort the page. Each 10x in referring domains gave 1.20x the odds of a top-3 spot, down from 1.51x.
- Position still matters for AI Overviews. They took their first source from page 1 86% of the time on long questions, and results at position 7 or lower were cited less (0.61x).
07AI Overview scorecard
Scorecard 1: the AI Overview ranking factors we tested
This is the main test. Each page is compared only with the other pages on its own results page, with all traits counted at once, across all 400 questions. Green helps, red hurts, grey could be chance. Backlinks land on the wrong side of 1.
3,140 page-1 results on 383 questions where at least one result was cited and one was not.
| Trait | Short questions | Long questions |
|---|---|---|
| 10x more referring domains | 0.85 (not sig.) | 0.79 |
| Product or landing page | 3.72 | 2.70 |
| Video | 3.18 | 4.20 |
| Brand's own site, brand named in question | 1.81 (not sig.) | 2.13 |
| Number in the title | 0.25 | 0.48 |
| Year in the title | 0.54 | 0.76 |
| Forum thread | 0.21 | 0.68 (not sig.) |
| Ranks for twice as many questions | 1.38 | 1.13 |
- Bigger sites look better only on the surface. Raw cited rates by referring domains: 44% under 1k, 40% for 1k to 10k, 55% for 10k to 100k, 60% above 100k. The top group is mostly Reddit and YouTube. Compare pages on the same results page and the backlink effect turns slightly negative.
- On long questions the backlink result is clearer. 0.79x per 10x, against 0.85x on short questions.
08Freshness
Content freshness: newer pages lead, more so on long questions
72%
of AI Overview sources on long questions showed a date (vs 25% on short)
130
days: median age of dated sources on long questions (vs 178 on short)
60%
of dated sources on long questions were under 6 months old (vs 50%)
14%
were over a year old (vs 26% on short questions)
238 dated sources on short questions, 743 on long. Age measured on 10 November 2026.
- Newer pages get the top spots inside the answer. On long questions the first 3 sources had a median age of 117 days and later sources 149 days. Short questions showed the same pattern (152 vs 232).
- Small sites that got cited showed their date. On long questions 83% of cited small-site pages carried a date, vs 57% for bigger sites. Their dated pages were about as fresh as big sites' pages (median 131 vs 126 days), so the date itself is not a small-site advantage. Showing it is.
- Freshness matters most for buying questions. Median age on long questions: 100 days for pricing and "vs", 107 for "best X", 132 for alternatives, 210 for how-to guides.
- A new year in the title is not freshness. 28% of cited pages titled "2026" carried a date over 6 months old. Titles with a year were cited 33% of the time vs 56% without.
- Old pages are not shut out. 14% of dated sources on long questions were over a year old, and evergreen how-to guides still get cited.
What we can and cannot say: Google shows dates on AI Overview source cards, not on most page-1 results. So we can say cited pages skew recent, and more so on long questions. We cannot compare them with the dates of pages that were not cited.
09Titles & page formats
Listicles get ranked. They do not get quoted.
3,194 page-1 results on questions with a usable AI Overview.
Title or page has itDoes not
3,194 page-1 results. Raw rates; section 07 has the same-page test.
- The number in the title is the tell. "11 Best X" titles: 0.37x the odds of being cited on the same results page. Titles with both a number and a year were cited 19% of the time vs 61% for titles with neither.
- Question-style titles look strong in raw numbers only because how-to questions get cited more. On the same results page they made no difference.
10Where the page lives
Where a page lives on the site changes its odds
We grouped page-1 results by the section of the site they sit in, using Google's breadcrumb. This part uses the 200 short questions.
1,025 page-1 results in the sections shown.
1,503 page-1 results on 190 short questions. The /compare/ row compares comparison pages with each other only.
11ChatGPT scorecard
Scorecard 2: the ChatGPT ranking factors we tested
ChatGPT plays a different game. It links almost only to company sites, mostly homepages and pricing pages, and here backlinks do predict who gets the link: 3.64x the odds per 10x more referring domains.
1,901 page-1 results on 231 questions where ChatGPT linked at least one page-1 site. Videos were never linked, so they are left out.
| Referring domains | AI Overview: page-1 results cited | ChatGPT: page-1 company sites linked |
|---|---|---|
| Under 1,000 | 44% | 4% |
| 1,000 to 10,000 | 40% | 28% |
| 10,000 to 100,000 | 55% | 32% |
| Over 100,000 | 60% (mostly Reddit and YouTube) | 46% (57 sites) |
Google page 1: 3,266 results. AI Overviews: 2,309 sources. ChatGPT: 372 answers, 1,592 links.
- It links to the vendor, not the article. On long questions 50% of ChatGPT's links went to homepages, 23% to pricing pages and only 7% to blog posts or guides.
- Small vendors are mostly left out. A company site under 1,000 referring domains that ranked on page 1 was linked 4% of the time.
- Long questions open a small gap. 28% of ChatGPT's links on long questions went to small sites, vs 20% on short ones.
- Little overlap with Google. Only 21% of ChatGPT's links on long questions were on Google's page 1, and 16% were AI Overview sources.
- A few brands take a large share. Across both rounds HubSpot was linked 61 times, Salesforce 37, Zoho 32.
12Side by side
AI search ranking factors, side by side
| Trait (same results page) | AI Overviews | ChatGPT |
|---|---|---|
| 10x more referring domains | 0.80 | 3.64 |
| Product or landing page | 3.37 | 4.79 |
| Brand's own site, brand named in question | 1.98 | 3.85 |
| Ranks for twice as many questions | 1.24 | 0.75 |
| Number in the title | 0.37 | 0.57 |
| Year in the title | 0.65 | 0.93 (not sig.) |
| Forum thread | 0.42 | 0.02 |
| Review site or directory | 0.77 (not sig.) | 0.09 |
| Video | 3.15 | never linked |
| Position 7 or lower | 0.61 | 0.43 |
| Fresh page | newer pages lead | not measurable |
AI Overviews reward
The right page, recently updated
- Product, pricing and comparison pages that answer the exact question
- Pages that show a real, recent date
- Short videos
- Not: backlinks, numbered list titles, a year in the title
ChatGPT rewards
An established vendor brand
- The vendor's own homepage and pricing page
- Sites with more referring domains
- The brand the buyer named
- Not: forums, review sites, blog posts
"Not measurable" for ChatGPT: it mostly links homepages and pricing pages, which rarely show a date.
13What to do
Actions, in order of payoff
- Write for the long question.Build pages around a full buyer situation (team size, budget, tools, use case), not a two-word keyword. On long questions small sites took half the top 3 and most AI Overview sources.Top 3: 31% to 50% · AI sources: 38% to 58%
- Match the page to the intent.Learning questions: guides. Shortlist questions: product pages and videos. "Alternatives": comparison pages and videos, since the named brand is rarely cited. "X vs Y" and pricing: your own comparison and pricing pages, because AI Overviews cite the vendor's site.Small-site gain: learn +30 pts · shortlist +20 · decide +8
- Keep money pages fresh, and show the date.Update "best X", pricing and comparison pages every few months with real changes, and show the updated date.72% of long-question sources dated · median 130 days · first 3 sources 117 days
- Fix your pricing, product and homepage before writing another list post.They win on both engines.Product page 3.37x (AI Overviews), 4.79x (ChatGPT) · ChatGPT links: 50% homepages, 23% pricing
- Drop the number from list titles."Best CRM for small B2B teams" over "11 Best CRMs for Small B2B Teams in 2026".Number in title 0.37x · number and year 19% vs 61% cited
- Do not fake freshness with the year.Changing the year without updating the page does not change the date Google shows.Year in title 0.65x · 28% of cited "2026" pages over 6 months old
- Move comparison pages out of the blog.Give them a /compare/ or /vs/ section, one page per competitor.69% vs 30% cited · 3.6x the odds (short questions)
- Make a short video for each money question.Videos are cited far more often than they rank.Video 3.15x · in 78% of short and 40% of long AI Overviews
- For ChatGPT, build the brand.Links, mentions and a clear vendor homepage and pricing page. This is the one place backlinks showed up.Backlinks 3.64x per 10x · small company sites linked 4%
- Stop buying links for AI Overviews.Spend that budget on pages and updates.Backlinks 0.80x per 10x across 400 questions
14Limitations & scope
What this study can and cannot say
- One collection window. Data from 1 to 11 November 2026, US English results, signed out. AI answers change over time.
- B2B SaaS only. Consumer, local and news questions may behave differently.
- Our long questions. We wrote them to match how people describe their situation to AI tools. Real long queries vary more.
- Site-level backlinks. Counts come from a free public index and cover the whole site, not the page. Paid tools may give different counts.
- Correlation, not cause. Comparing pages on the same results page removes topic effects, but traits we did not measure may explain part of what we see.
- Not measured: schema markup, word count, headings, author details and other on-page structure.
- Dates. Only what Google shows on AI Overview source cards. We cannot compare cited pages with uncited ones on age.
- ChatGPT. 372 of 400 answers (free-tier limits), signed out, links only.
- Labels. Page type and site section come from title and URL rules, so some will be wrong.
15FAQ
Common questions
Are backlinks still important?
It depends on where you want to show up. For Google AI Overviews, no: on the same results page, 10x more backlinks gave 0.80x the odds of a citation. For Google rank, a little on short questions and less on long ones. For ChatGPT, yes: 10x more backlinks gave 3.64x the odds of a link.
How do you rank in AI Overviews?
Answer the full question on the right kind of page, and keep it current. Product and landing pages had 3.37x the odds of being cited, videos 3.15x, and 72% of sources on long questions showed a date. More backlinks, a number in the title and a year in the title were all linked to lower odds.
How do you get cited by ChatGPT?
Be an established vendor with a clear homepage and pricing page. On long questions 50% of ChatGPT's links went to homepages and 23% to pricing pages, and sites with 10x more referring domains had 3.64x the odds of a link.
Why do long questions favour small sites?
A long question names a situation, and pages written for that exact situation match it better. Those pages are often on smaller, newer sites. The data shows the shift clearly; the reason is our reading.
Does content freshness matter for SEO and AI search?
For AI Overviews it looks that way. On long questions 41% of dated sources were under 3 months old and 60% under 6 months, and newer pages took the first spots in the answer. Pricing and "best X" pages lean newest; how-to guides can stay evergreen. We could not compare with the age of pages that were not cited.
Are long-tail keywords still worth targeting?
Yes. On long questions, sites with under 1,000 backlinks took 50% of Google's top 3, vs 31% on the short versions of the same questions, and only 24% of page-1 sites were the same for both.
Should I take the year out of my titles?
A year in the title was linked to lower citation odds (0.65x). Update the page for real and show the date instead.
Which questions should a small site go after?
Long learning and shortlist questions. There, small sites took about half of the top 3. On pricing and "X vs Y" questions the vendors and big comparison sites still hold most spots.
Does ChatGPT use Google's results?
Mostly not the same ones. About 21% of its links were on Google's page 1 and 16% were AI Overview sources.
Did you test schema markup?
No. That needs reading each page's code, which was outside this round. It is the natural next study.
16How to cite & about
How to cite this research
Suggested citation
https://www.growthspreeofficial.com/reports/the-long-question-shift
When citing a single figure, include the sample, for example: "On long buyer questions, sites with under 1,000 backlinks took 50% of Google's top 3 (400 B2B SaaS questions, US) (Manchanda / GrowthSpree, 2026)."
About the research
An independent study of how backlinks, question length, freshness, page type, titles and URL structure relate to Google rankings and AI citations. All data was collected by hand in a regular browser from free, public sources: Google results, Google AI Overviews, ChatGPT answers and a public backlink index. No paid SEO tools or APIs were used. The report and the full dataset are free to download.