Quick answer: To rank in AI search and Google AI Overviews, focus on the signals these systems can actually read: a clear, answer-first page, clean structure, genuine topical authority, and a site that AI crawlers can reach and parse. There is no ranking dial to turn. Inclusion correlates with clarity, credibility, and crawlability, but the selection logic is proprietary and still shifting, so treat any tactic as a bet rather than a guarantee.
"Ranking" is not quite the right word for what happens in AI search, and that mismatch trips people up. In classic search you occupy a numbered slot. In AI Overviews and engines like Perplexity, a model reads a handful of sources and writes one answer, citing some of them. You are not ranking number three. You are either in the answer or you are not.
So the real question is: what makes a page one of the sources a model chooses to read and cite? Here is what we can say honestly, and what we cannot.
How AI surfaces appear to pick sources
Nobody outside Google, OpenAI, and Perplexity has the selection algorithm. But the observable behavior across these systems is consistent enough to describe.
They start from a retrieval step. The engine runs one or more searches, often several reformulated queries for a single user question, and pulls a candidate set of pages. Then it reads those candidates and synthesizes an answer, attaching citations to the sources it leaned on. That means two separate gates: your page has to surface in the retrieval step, and then it has to be clear and credible enough to survive the synthesis step.
A page can pass the first gate and fail the second. It ranks, it gets retrieved, and then the model reads three paragraphs of throat-clearing, gives up, and quotes a cleaner competitor. AI search punishes buried answers harder than classic search does.
What correlates with inclusion
We are careful with the word "correlates" on purpose. These are the properties shared by pages that tend to get cited, based on how the systems behave, not a leaked ranking formula.
| Signal | Why it plausibly matters | What you can do |
|---|---|---|
| Answer clarity | The model needs a clean claim to quote | Lead with the answer, one idea per section |
| Structure | Chunked, well-headed pages retrieve cleanly | Descriptive H2s, tables, short lists |
| Topical authority | Engines favor sources that look credible | Deep coverage of a topic, real expertise |
| Freshness | Many queries prefer current sources | Update and re-date important pages |
| Crawlability | If it cannot be read, it cannot be cited | Server-rendered HTML, no needless walls |
| External corroboration | Being referenced elsewhere signals trust | Earn mentions and links from credible sites |
None of these is a secret. They are the same things that made content good before AI search existed. That is the reassuring part: the fundamentals did not get replaced, they got re-weighted.
A practical checklist
Work through this for any page you want cited in AI Overviews or AI search.
- Front-load the answer. The first paragraph should answer the query in plain language, no preamble. If a reader (or a model) leaves after two sentences, they should have the answer.
- Match headings to real questions. Use the phrasing people actually type. "How much does X cost" beats "Pricing considerations."
- Break the page into self-contained chunks. Each section should make sense pulled out on its own. Retrieval works on passages, not whole pages.
- Add structure a machine can parse. Tables, ordered lists, and schema markup all help systems extract facts cleanly.
- Keep it current. Put a visible date on it and refresh the substance, not just the timestamp.
- Verify crawlability. Confirm your answer is in the raw HTML, not injected by client-side JavaScript that a crawler may skip.
- Build real authority. Cover your topic thoroughly enough that a model has reason to treat you as a primary source.
If you want to see what the engines are actually retrieving for your target questions, you can run those queries through a search API that returns full page content and inspect the sources directly, rather than guessing from the answer alone.
Test crawlability the way a crawler sees it
The most common reason a good page never gets cited is that the answer is not in the HTML a crawler receives. If your content renders only after JavaScript runs, some AI crawlers never see it.
You can check this the same way an AI system would, by fetching the raw page and looking at what actually comes back:
curl https://link.sc/v1/fetch \
-H "Authorization: Bearer lsc_your_key" \
-H "Content-Type: application/json" \
-d '{
"url": "https://yoursite.com/your-answer-page",
"format": "markdown"
}'
If your answer is missing or mangled in the returned markdown, an AI crawler is probably seeing the same broken version. That is a fixable problem, and fixing it does more for AI visibility than most "optimization" tactics.
Be honest about the uncertainty
Here is the part most guides skip. AI search is early and opaque. Google has said little about exactly how AI Overviews select sources, and it has changed the feature repeatedly. Perplexity and ChatGPT are similarly closed. Correlation studies exist, but the systems shift under them, and anyone claiming a precise formula is overselling.
We wrote a whole piece on why this matters and how to reason about it, called AI search is a black box. The short version: you cannot reverse-engineer a moving, proprietary target, so you optimize for the properties that are stable across every version of these systems and stay skeptical of anything more specific.
What to actually do this quarter
If you want a plan rather than a philosophy:
- Pick your ten most important queries.
- Ask each of the major AI engines those queries and record who gets cited.
- Fetch the cited pages, compare their structure and clarity to yours.
- Rewrite your pages answer-first, with clean structure and current information.
- Confirm each page's answer is present in the raw HTML.
- Re-check citations in a few weeks and iterate.
That loop is boring and it works better than chasing whatever tactic is trending this month. The engines reward clarity, credibility, and crawlability. Build for those, measure what you can, and hold your conclusions loosely.
AI search will keep changing. Clear, well-structured, trustworthy content will keep being the thing it rewards. Bet on that.
See exactly what AI crawlers see on your pages. Fetch any URL to clean markdown with link.sc and check whether your answer even makes it through.