AIO Ranking Factors: What Google AI Overviews Actually Weight

I keep seeing the same AIO ranking factors list pasted across a dozen agency blogs: authority, freshness, schema, E-E-A-T, structured content, backlinks, entities. Ten items, equal weight, no numbers. If you follow that list you will spend a quarter on schema and author bios and move nothing.

The published studies tell a different story. Two or three factors do most of the work, a few are gates you either pass or fail, and several popular ones show no measurable effect at all. Domain Rating, for one, did not predict a single citation in the most careful within-SERP test I could find.

This post ranks AIO ranking factors by what the evidence says they weigh. I am not running a proprietary citation study here, and I would not trust anyone who claims to have cracked Google’s system from the outside. What I did is pull the four largest public datasets, check the figures against the source pages, and reconcile where they disagree. Then I built a working order of operations from it, which I call the Five Weights Framework.

If you want the broader strategy first, start with our answer engine optimization guide. This post is narrower: given limited hours, which factors do you touch first?

Why Most AI Overview Ranking Factors Lists Mislead

A checklist treats every item as a binary: done or not done. Ranking systems do not work that way. Some signals decide whether you are even considered. Others decide which of the considered pages wins. Mixing them in one flat list hides the difference.

The bigger problem is that most lists mix three different units of measurement:

What is being measured	Example claim	Why it gets confused
Page citation	”This URL appeared in the AI Overview”	Depends on the query and the page
Brand mention	”This brand was named in the answer”	Depends on the whole web, not one page
Click or traffic effect	”CTR dropped after the overview appeared”	A separate question entirely

A study that says branded mentions correlate with AI visibility is not saying your blog post needs more backlinks. A study that says top 10 pages get most citations is not saying rank 1 beats rank 3. Keep the unit straight and most of the apparent contradictions disappear.

We covered the traffic side in Google AI Overviews for B2B. Here I only care about selection: why one page gets quoted and its neighbor does not.

What the Studies Actually Measured

Four datasets carry most of the weight in this post. I fetched each one rather than relying on secondhand summaries, and I note the limits.

Study	Sample	Date	Headline finding
Ahrefs, citation vs rank	1.9 million citations from 1 million AI Overviews	July 2025	76.1% of cited pages ranked in the top 10; 9.5% at positions 11 to 100; 14.4% outside the top 100
Surfer, 20 factor rubric	650,000 plus AI answers across ChatGPT, AI Overviews, AI Mode, Perplexity	April 2026	First paragraph factors dominate; almost 40% of citations come from the first 100 words
On-Page.ai, within-SERP test	370 page one results, 50 keywords, 10 verticals	July 2026	None of ten authority and demand signals predicted citation
Ahrefs, brand correlation	75,000 brands	May 2025	Branded web mentions correlated at 0.664; backlinks at 0.218

Three caveats before you quote any of these. The Ahrefs citation study looked at the top 3 most visible citations per response, not every link. The On-Page.ai sample is small, 370 pages, so a real small effect could hide in the noise. And the Ahrefs brand study is correlation only; the authors say so themselves, and every coefficient in it sits in the moderate to weak range.

Surfer sells content tooling and On-Page.ai sells measurement, so each has a reason to like its own framing. I treat them as evidence, not verdicts. Where two sources agree from different methods, I lean on that. Where they conflict, I say so.

The Five Weights Framework

Here is the order of operations I use. Weights are relative and directional; I am not pretending these are Google’s coefficients.

Weight	Factor	Role	Evidence strength
1	Ranking eligibility	Gate	Strong: 76.1% of citations from the top 10
2	Opening answer	Primary selector	Strong: three largest weights in the Surfer rubric
3	Extractable passages	Selector	Moderate: semantic triplets weighted 11%
4	Brand and entity presence	Off-site amplifier	Moderate: correlation only, brand level
5	Freshness and hygiene	Tie breaker	Weak to moderate: directional

Notice what is missing: raw domain authority. I will get to that.

Weight 1: Ranking Eligibility Is the Gate

Start with the uncomfortable part. In the Ahrefs sample, 76.1% of cited pages ranked in the top 10, and the median rank of the most prominent citation was position 2. Surfer’s April 2026 data points the same direction: the top 3 positions took 80% of citations, with position 1 at 37.5% and position 2 at 31.8%.

Put plainly, if you are not on page one, no amount of passage polish will carry you. The 14.4% that came from outside the top 100 shows exceptions exist. But a plan that depends on exceptions is not a plan.

There is a wrinkle. Ahrefs found separately that only about 12% of citations from ChatGPT, Gemini and Copilot appeared in Google’s top 10 for the same query, because those assistants run query fan-out, retrieving across many rephrasings of the prompt. Perplexity sat higher at roughly 28.6%. Google’s own overviews behave differently: they lean on the live ranking. So advice that works for ChatGPT citations does not transfer cleanly to AI Overviews. Anyone who sells one playbook for every engine is blurring this.

What to do with it: build a simple eligibility table before touching content.

Your page’s position for the target query	Action
1 to 3	Optimize the opening and passages; you are already the likeliest source
4 to 10	Optimize the opening first; this is the cheapest win band
11 to 30	Treat as a ranking project; AIO tuning alone will not rescue it
Beyond 30	Skip AIO tuning; fix intent match and topical coverage first

The 4 to 10 band is where I would start. These pages already pass the gate and usually have the weakest openings, because they were written for a headline and not for extraction.

Weight 2: The Opening Answer Is the Primary Selector

This is where the data gets striking. In Surfer’s rubric, the three heaviest factors all measured the first paragraph:

• Early query confirmation, 22.3%: does the first paragraph signal that you are answering the question asked?
• Search intent alignment, 22.2%: does the content address why the person searched, not just the literal words?
• Early query answer, 18.9%: do you give the short version before you start elaborating?

That is 63.4% of the rubric’s predictive weight sitting in one paragraph. And almost 40% of all citations came from the first 100 words of a page.

I will be honest about how to read that. A rubric weight is not a Google weight; it measures how well each factor separated cited pages from uncited ones in Surfer’s sample. But the direction matches what I see when I audit pages. The pages that get quoted answer first. The pages that get skipped open with a story, a statistic about the market, or three sentences of context.

A rewrite test for your openings

Take your target page and answer three questions about its first 100 words:

1. Could a stranger quote this paragraph alone and be correct? If the answer needs the third paragraph, it fails.
2. Does it name the thing being asked about, in the words the searcher used? “Our approach to retention” fails. “Churn prediction uses three signals” passes.
3. Does it commit? Hedged openings (“it depends on many factors”) give the model nothing to lift.

Here is the shape of a before and after.

Version	Opening
Before	”In today’s market, SaaS teams are under pressure to understand churn. There are many approaches, and the right one depends on your stage.”
After	”Churn prediction works best with three signals: usage decline over 14 days, support ticket sentiment, and billing events. Start with usage decline, since it is the earliest of the three.”

The second version is shorter, commits to an answer, and contains a quotable claim. Do this for every page in the 4 to 10 band. It is an hour of work per page and it is the highest weighted change on the list.

Weight 3: Extractable Passages Decide the Rest

The fourth factor in Surfer’s rubric was what they call semantic triplets, weighted at 11%: complete statements a model can quote directly. A triplet is a subject, a relationship and a claim, packaged so it stands alone.

Compare: “This improves outcomes significantly” (cannot be quoted, no subject) against “Answer first openings gave cited pages the largest share of rubric weight in Surfer’s April 2026 study” (can be quoted, self-contained).

One more data point worth knowing: an analysis cited in several GEO guides found AI systems read a median of roughly 377 words per page. I could not trace that figure to a primary study, so I would not build a strategy on it. But it fits the finding that citations cluster at the top, and it is a good reason to make every section open with its answer, not just the page.

Practical rules I apply when rewriting passages:

• Open every H2 with the answer. Two sentences, then the explanation.
• Put the number and its source in the same sentence. A claim and its evidence split across paragraphs cannot be extracted together.
• Use real tables for comparisons. They give the model a structured lift. Our AEO audit checklist for citation readiness covers the audit side.
• Define terms inline. If a passage only makes sense after reading the page before it, it will not be chosen.

Surfer’s bottom performers are instructive here. Author bios scored 5.8%, tables of contents 3.7%, and FAQ sections 2.6%. FAQ blocks are the most recommended AIO tactic on the internet, and in this rubric they were close to the least predictive. They are not useless; they are just not where your first hours should go.

Weight 4: Brand and Entity Presence Works at the Brand Level

Now the off-site factors. Ahrefs measured 75,000 brands with Domain Rating above 40 and counted how often each appeared in AI Overview responses for its highest volume keyword. The correlations, in Spearman terms:

Signal	Correlation with AI Overview brand mentions
Branded web mentions	0.664
Branded anchors	0.527
Branded search volume	0.392
Domain Rating	0.326
Referring domains	0.295
Backlinks	0.218
URL Rating	0.180

Branded mentions beat backlinks by roughly three to one. That is the number you see quoted everywhere. Here is what the quotes leave out. About 26% of brands showed zero mentions and were excluded. The study unit is the brand, not the page. And the authors flag correlation, not causation: brands that are mentioned a lot are probably also big, searched for, and linked, so the mention count may be a proxy for being a well known company.

What I take from it: if you are a small B2B SaaS brand with no third party mentions, a perfect page is competing against pages from brands that the web already talks about. That is a long game. It runs through category research, podcast and newsletter placements, customer review sites and communities, not through more on-page tweaks. Our piece on entity SEO for AI search covers the mechanics.

Keep expectations honest. You will not close a brand gap in a month, and the page level weights above are what you can change this week.

Weight 5: Freshness, Schema and Other Hygiene

This is the weakest evidence tier and I want to be careful, because this is where most blog posts invent numbers.

Freshness. I have seen claims that pages updated within 30 days earn about three times the citations. I could not verify that against a primary study, so I am not repeating it as fact. What I am comfortable saying: for queries where recency matters (pricing, tool comparisons, regulations, benchmarks), an out of date page is a weaker candidate, and changing the visible date without changing content does not fix it. Update the substance and the numbers.

Schema markup. Google’s own guidance says no special markup is needed to appear in AI features. Third party reports are mixed: some correlate schema with citation, while at least one matched control test I came across found no reliable lift. I treat schema as hygiene. Add Article, Organization and FAQPage where they honestly describe the page. Do not expect it to reorder anything.

Domain authority. This is the finding that should change budgets. In On-Page.ai’s July 2026 test, Domain Rating, URL Rating, page backlinks, referring domains, organic traffic, keyword count, brand search volume, page age and keyword match domains all failed to predict citation within page one results. None of the ten reached significance.

How does that fit with Ahrefs showing Domain Rating at 0.326 in the brand study? Different questions. Authority helps you rank, which gets you through the gate. Once five or ten pages have passed the gate, authority no longer separates the winner. That is consistent with both studies, and it is the single most useful sentence in this post for budget decisions.

Factor	Gets you into the pool	Wins inside the pool
Domain authority and backlinks	Yes, through ranking	No evidence
Opening answer quality	Indirectly	Yes, strongest evidence
Passage extractability	No	Yes, moderate
Brand mentions	Possibly	Unclear, correlation only
Freshness	Query dependent	Tie breaker
Schema	No	No reliable evidence

A 30 Day Plan Using the Weights

Here is how I would sequence a month for a B2B SaaS site with 30 to 100 posts.

Week 1: Build the eligibility table

Export your target queries and current positions from Search Console. Mark each page as 1 to 3, 4 to 10, 11 to 30, or beyond. Pick the 4 to 10 band and shortlist 10 pages by commercial value. Check which of them already trigger an AI Overview, since not every query does.

Week 2: Rewrite the first 100 words

Run the three question test on each shortlisted page. Rewrite openings so they confirm the query, match the intent and answer in two sentences. Keep a before and after log with the date so you can attribute changes later.

Week 3: Restructure the passages

Open each H2 with its answer. Move the statistic next to its source. Convert one comparison per page into a table. Add schema that truthfully matches the page, and move on.

Week 4: Measure and start the slow work

Re-check which of the 10 pages now appear as cited sources. Run the same prompts a few times, because overviews vary between runs, and record the share, not a single yes or no. Start one off-site workstream in parallel: two or three genuine third party placements a month. If you need a structured way to track this, see our write up on AI citation monitoring and the broader AIO optimization framework.

Be realistic about outcomes. In my experience a page that already ranks 4 to 10 and gets a proper opening rewrite is the most likely candidate to show a change inside a month. I do not have a clean number for the lift, and I would be suspicious of anyone who does.

Mistakes I See Teams Make

1. Spending the quarter on schema. It is the cheapest thing to ship, so it gets shipped first. It is also the weakest evidence on the list.
2. Chasing backlinks for AIO specifically. Links help you rank. They do not appear to pick the winner among ranked pages.
3. Optimizing pages that rank on page four. The gate does not open for them. Fix ranking, then fix extraction.
4. Writing openings for humans only. A story opening is fine for readers who arrive already interested. Overviews skip it.
5. Treating all engines as one. Overviews lean on live rankings. ChatGPT and Gemini lean on fan-out retrieval. Different mechanics, different priorities.
6. Trusting a single study. The citation overlap figure has moved between studies and samples. Look for two methods agreeing before you reorganize a roadmap.

What to Do on Monday

Pull your 10 best commercial pages that rank 4 to 10. Read only their first 100 words. If a stranger could not quote that paragraph and be right, rewrite it before you touch anything else. That one change targets the factors with the heaviest published weight and costs almost nothing.

If you want a second pair of eyes on which pages are worth the effort, book a free growth audit and we will map your AI Overview exposure against your ranking positions. You can also try our free AI growth tools at app.momentumnexus.com.