Do Backlinks Still Matter for AI Visibility?

Yes — when the backlink sits on a third-party URL an answer engine can open, quote, and treat as evidence. No — when the href exists only to juice a Domain Rating number nobody in ChatGPT, Perplexity, or a Google AI Overview will ever retrieve. The object that still matters is the evidence page. The object that died is the referring-domain shopping list.

I am William Spurlock, founder, AI Systems Architect, and Fractional AI CTO. I have been SEO-certified since 2021; the work now sits under AEO, AIO, and GEO. I have shipped 600+ automations with 500+ live, logged 20,000+ hours inside agentic systems, and deleted 35,000+ hours of client busywork across that book of work. I do not invent citation percentages, fake "X% of AI answers come from backlinks," or client names you cannot verify.

This spoke owns one question: Do backlinks still matter for AI visibility? The June 19 map of
where SEO and AI visibility overlap and where they split
owns the wider fork. The June 11
digital PR for AI visibility
post owns how mentions get earned. This page is the keep-versus-drop for the URL itself — when a third-party link still helps an answer engine, and when it is SEO theater.

If you want the owner playbook for getting named in chat, that lives in
how to get ChatGPT and Perplexity to recommend your business.
If the brand node itself is fuzzy, start with
entity SEO.
Do not mix those jobs into a link-building campaign.

---

Do backlinks still matter for AI visibility? #

They still matter as third-party evidence URLs and as one input into Google's ranking pool. They do not matter as a raw backlink count, a bought Domain Authority package, or an exact-match anchor ratio. I treat a backlink as useful only when a retrieval system can land on that page and lift a fact, a name, or a source chip from it.

Three objects get mixed together in almost every "links are dead / links still rule" thread. Separate them or you will optimize the wrong one.

Object	What it is	What an answer engine does with it
Backlink	An href on someone else's page that points at you	Google may use it as a quality hint for ranking. ChatGPT and Perplexity do not score your Ahrefs report.
AI citation	Your brand name or a URL of yours appears as a source inside the answer	That is the visibility unit. A ranking under a Google AI Overview with no source chip is not this.
Evidence URL	A third-party page a human could open and learn a specific fact from	Retrieval systems can fetch it, quote it, and show it as a source — whether or not the href points at your homepage.

Google still says, in public, that other prominent sites linking or referring to a page is one quality signal. That sentence lives on How Search works / ranking results. Notice the wording: link or refer. A named mention on a trusted page is already in Google's own quality language. The href is not the only receipt.

Google AI Overviews are described by Google as an AI-generated snapshot that can include links to supporting pages (AI Overviews help). The supporting page is the evidence. Your referring-domain graph is not on that snapshot.

The dated measurements I will use on this page — and the ones I refuse to invent — sit here:

Source	Date	What it actually measured	What it does not say
Ahrefs, 75k brands	May 26, 2025	Branded web mentions correlated 0.664 with AI Overview brand visibility. Raw backlink count correlated 0.218. Domain Rating correlated 0.326. Ahrefs states correlation is not causation.	It does not say "X% of citations come from backlinks."
Ahrefs, ChatGPT / AI Mode / AI Overviews	December 12, 2025	Branded web mentions stayed in the 0.66–0.71 range. Link-volume metrics stayed weak across ChatGPT, AI Mode, and AI Overviews. YouTube mentions led at about 0.737.	It does not give you a safe backlink quota.
Ahrefs, AI Overview citations vs top 10	March 2026 update	37.9% of URLs cited in AI Overviews also sat in the first 10 organic blocks, down from about 76% in their July 2025 cut.	Rank is a weak proxy for a citation. A #3 blue link is not a source chip.
SparkToro / Similarweb	Jan–Apr 2026 window, published 2026	68.01% of U.S. Google searches ended with no click. SparkToro puts AI Overviews on 20%+ of searches and says those queries lose nearly 60% of CTR when an Overview is present. Methodology changed vs their 2024 Datos study (58.5% U.S. zero-click).	Zero-click is not a backlink statistic. It is why a citation can matter more than a click you never received.
Seer Interactive, AIO CTR v3	Analysis April 2026; data through Feb 2026	Cited brands earned more clicks per impression than uncited brands on Overview-present queries. Seer will not call that proof of cause.	This is a citation-versus-uncited gap, not a link-graph study.

My working opinion, held loosely and tested against those tables: I would rather have five third-party pages a Perplexity source list can open than five hundred referring domains in a crawl report that no engine will fetch. Volume still moves classic rank at the margin. Volume does not mint an evidence URL.

That is the whole thesis. The rest of this post is the test for "still helps" versus "theater," then the keep-versus-drop, then how you check the work without buying another 'AI backlink' product.

---

When does a third-party URL still help an answer engine? #

A third-party URL still helps when a retrieval pass can use that page as a source, or when Google can use the link-or-refer signal to keep your own page in the ranking pool those Overviews still draw from. If neither thing is true, the href is decoration.

Answer engines do not share one scoring function. Google Search still runs a link-aware quality stack. Google AI Overviews sit on top of Search and attach source chips to a snapshot. ChatGPT with search and Perplexity retrieve live pages and show URLs. Perplexity's own Agent API docs treat citations as search_results items with a real url, title, and snippet — and they tell developers not to invent those URLs because models hallucinate them (streaming citations). The evidence has to already exist as a fetchable page.

I keep three help paths. Only the first one is "backlinks" in the 2016 sense.

1. The ranking-pool path (Google still uses links) #

Google's ranking explainer still names prominent sites that link or refer as a quality clue (How Search works). Google's spam policies still define link spam as creating links primarily to manipulate rankings — paid links that pass credit, excess exchanges, automated links, low-quality directories, footer/widget schemes, and guest-post or press-release links with optimized anchors.

So the honest version is:

• A real editorial link from a page Google already trusts can still help your URL rank.
• Ahrefs' March 2026 cut says only about 38% of AI Overview citations also sat in the organic top 10. Rank is an incomplete on-ramp, not a citation machine.
• Buying the link, templating it, or stuffing the anchor is the same policy violation it was before Overviews existed.

If your only theory is "more referring domains → more AI Overviews," the May 2025 Ahrefs table already disagrees with you. Mentions beat backlink count. Rank still matters some of the time. It is not the sport.

2. The retrieval path (they cite the other page) #

This is the path most SEO decks skip.

ChatGPT with search and Perplexity do not need your homepage to win for you to show up. They need some indexed page that states a usable fact: who you are, what you sell, where you work, what you measured. That page is often not yours. It is a trade write-up, a comparison table, a transcript, a directory record with a real editorial body, or a research note that names you in a sentence a model can lift.

When that happens, the "backlink" is optional. The URL of the third-party page is the asset. The engine cites them. Your brand rides in the prose.

That is why an unlinked mention on a page people actually read can outperform a followed href on a page nobody would send a buyer. Ahrefs said the same thing in plain language in May 2025: unlinked mentions do little for classic SEO and much more for GEO, because models read words, co-occurrence, and context (Ahrefs / Ryan Law, cited in the 75k study).

3. The entity-corroboration path (the page proves you exist) #

Answer engines are cheap skeptics. A claim that lives only on your domain is a claim without a witness. A second domain that repeats the same legal name, the same service, the same city, and the same person is corroboration. The href is a convenience. The repeated string is the signal.

That is entity SEO
work, not a referring-domain contest. If the third-party page says "Acme HVAC" and your site says "Acme Heating & Cooling LLC dba Ice Kings," the model has two nodes, not one brand.

When the URL still earns its keep #

A third-party URL still helps when most of these are true:

• A stranger can open it and learn one specific, dated fact about you.
• The page is indexed and not trapped behind a login, a soft-404, or a thin affiliate shell.
• Your name is written as a name, not as a stuffed exact-match keyword.
• The surrounding copy is about the topic a buyer would ask GPT-5.5, Gemini 3.1 Pro, Claude Opus 4.8, or Perplexity.
• You would not be embarrassed if that URL appeared as a source chip next to your brand.

A third-party URL stops helping when:

• The only unique token on the page is a footer href.
• The page exists to sell links, not to inform a reader.
• The anchor is a keyword you paid someone to type.
• The domain is an expired shell, a PBNish neighborhood, or a "submit URL" directory.
• You cannot explain, in one sentence, what fact the engine would extract.

Engine	What the third-party URL can do	What your backlink report cannot do
Google Search	Pass a quality hint if the link looks editorial; help discovery	Guarantee an AI Overview source chip
Google AI Overviews	Become a supporting link on the snapshot, or help your own URL stay in the pool the snapshot draws from	Turn a DA package into a citation
ChatGPT with search	Enter the retrieved set as a page the model can quote	Make GPT-5.5 "count" your referring domains
Perplexity	Appear in search_results as a real URL with title and snippet	Invent a source the index never fetched

I do not run a separate "AI link graph." I run an evidence-URL test. If the page cannot survive that table, I do not care what the crawl tool scored it.

---

When is a backlink just SEO theater? #

A backlink is SEO theater when it was built to move a third-party authority metric and cannot serve as a page an answer engine would retrieve. Theater looks busy in Ahrefs. It does not show up as a source chip.

Google already published the script. The link spam section names the moves: paying for links that pass ranking credit, goods-for-links, forced-link contracts, excess exchanges, automated insertion, low-quality directory and bookmark links, keyword-stuffed widgets, sitewide footer and template links, optimized forum signatures, and low-value pages created to manufacture linking signals. Guest posts and press releases with optimized anchors are on that list. That is not my opinion. That is Google's current spam policy page.

Theater has a second tell: the metric you are celebrating is the weak correlation.

On May 26, 2025, Ahrefs put raw backlink count at 0.218 against AI Overview brand visibility, referring domains at 0.295, and Domain Rating at 0.326. Mentions sat at 0.664. The December 12, 2025 rerun said link-volume metrics stayed weak across ChatGPT, AI Mode, and AI Overviews, and that site-page count sat near 0.194. Ahrefs repeats the hedge both times: correlation is not causation. I will repeat it too. What I will not do is treat a 0.218 relationship as a reason to buy another hundred links.

Theater I see on actual sites, not a hypothetical agency roast:

Move	What the deck claims	What I see in the answer
DA / DR shopping	"We got you 20 DR-50 links"	No engine publishes a DR threshold. Ahrefs' own DR correlation was mid-pack and weaker than mentions.
Exact-match guest posts	"We built topical authority"	Google lists optimized anchors in guest posts as link spam. The page is rarely a page a buyer would read.
Directory dumps	"Citations for local + AI"	A NAP row with no prose is a database field, not an evidence paragraph. Thin directories are on Google's spam list.
Footer / widget / profile blasts	"Sitewide authority"	Sitewide template links are named in the same policy. Retrieval does not want your footer.
Expired-domain networks	"Aged authority"	You bought a corpse. The content is not a witness. Google's expired-domain and site-reputation abuse rules exist because this pattern kept coming back.
Homepage-only vanity links	"Brand links"	The homepage says you exist. It rarely answers the buyer question the Overview or the chat is synthesizing.
Paid placements marked as editorial	"Digital PR"	If money changed hands and the href passes credit, Google wants rel="sponsored" or nofollow. An unmarked paid link is spam, not evidence.
Anchor-ratio dashboards	"Natural profile"	Answer engines read sentences. They do not grade your 15% exact-match pie chart.

The theater test I use on a Monday, without a new tool:

1. Open the linking URL in an incognito window.
2. Scroll until you find the sentence that names you.
3. Ask: would I send this URL to a founder who asked "should I hire these people?"
4. If the honest answer is no, it is theater. Log it as hygiene, not as AI visibility progress.

I have sat through reports where the referring-domain count went up every month and the brand still never appeared as a source in Google AI Overviews, ChatGPT, or Perplexity. The report was not lying. It was measuring the wrong object. SparkToro's 2026 zero-click cut is the mood music here: most Google searches already end without a click. A link that only exists to manufacture a click you were not going to get is a costume.

One more tell. If the vendor cannot show you the third-party URL — the live page, not a screenshot of a metric — they are selling theater. Evidence has a URL. Theater has a dashboard.

I will not write you a guest-post sequence or a digital-PR pitch list on this page. That is a different job, and it already has a home. This page's job is to keep you from paying for the costume.

How to read a vendor report without getting sold theater #

Most "AI visibility + links" decks I am handed have four slides that look like progress and one slide that is the actual job. I read them in this order, and I stop the meeting if slide four is missing.

Slide they lead with	What I ask	If they cannot answer
Referring domains this month	Name the five URLs, not the count	The campaign is a volume buy
Domain Rating / Domain Authority gained	Show me a page Google or Perplexity would open	They sold you a third-party score
"Natural" anchor mix	Show me the sentence the anchor sits in	The mix is a costume for exact-match work
Indexed, dated pages that name you	Open them with me, live	There is no evidence layer
Citations in Overviews or chat	Which query, which engine, which source URL	They are reporting SEO and calling it AEO

I have ended more than one of those meetings at the third row. That is not rudeness. That is refusing to pay for a metric Ahrefs already put at 0.218 against the thing you said you wanted.

If you are the operator writing the brief, invert the deck. Start with the ten buyer questions. Require a witness URL per question or an honest "none." Then, and only then, talk about whether an href on that witness page is even required. Most of the time the paragraph was the win and the vendor is upselling the click.

---

What should you keep versus drop? #

Keep third-party pages that can stand as evidence. Drop campaigns whose only output is a followed href and a prettier crawl chart. If a tactic cannot produce a URL you would let Perplexity put under your name, it does not belong on the AI-visibility list.

This is a filter, not an acquisition playbook. How you earn a real mention is the
digital PR
spoke. How you become a recommendable entity is the
ChatGPT and Perplexity
pillar. How SEO and AI visibility share DNA is the
overlap
map. Stay on the filter.

Keep #

Editorial pages a stranger would finish. A trade article, a comparison with a real table, a research note with a date, a transcript with show notes that state a fact. The href is a bonus. The paragraph is the asset.

Brand-name mentions, linked or not, on topical pages. Ahrefs' May 2025 and December 2025 tables both put branded web mentions ahead of backlink volume. Branded anchors (your name as the clickable text) sat in the middle — 0.527 in May, and up to 0.628 on AI Mode in December. If someone is going to link, the brand name is the honest anchor. "Best [keyword] agency" is the 2018 leftover.

Your own extractable pages. When an Overview or a chat does pick you, it still needs a URL it can quote. Question-led headings, first-sentence answers, FAQ blocks, consistent Organization names. A beautiful backlink into a slogan homepage is a door into an empty room.

Qualified paid links, labeled as paid. Ads are allowed. Google says buying links that pass ranking credit is not, unless you qualify them with rel="sponsored" or nofollow (qualify outbound links). Keep the sponsorship. Drop the pretend-editorial.

Search Console's external-links export as a hygiene list. Use it to find spam, expired junk, and sitewide templates you should disavow only if Google has a manual action or a clear pattern. Do not use it as an AI-visibility scoreboard. The Links report does not tell you whether Gemini 3.1 Pro or GPT-5.5 retrieved the page.

A short list of witness URLs you can name from memory. If you cannot name five third-party pages that state a specific fact about you, you do not have an evidence layer. You have a site and a hope.

Drop #

Referring-domain volume targets. "We need 80 new domains this quarter" is a ranking-era KPI. Ahrefs already showed the count is the weak number.

Exact-match anchor campaigns. Google's spam policy calls this out in guest posts and press releases. Answer engines do not need the keyword in the blue text. They need the fact in the sentence.

Directory farms and bookmark sites. Google lists them. A NAP-only row does not give a model a paragraph to lift. A real local listing with a written profile can; a submit-and-forget farm cannot.

Expired-domain and private-blog-network neighborhoods. You are buying history that is not yours. That is not corroboration. It is costume jewelry with a WHOIS scar.

Homepage-only vanity acquisition. Unless the homepage answers the buyer question, you are collecting hrefs to a billboard. Put the rare editorial link on the page that holds the answer.

Unmarked paid placements sold as "features." If you paid, qualify the link. If the publisher will not qualify it, you are buying spam, not evidence.

Anchor-ratio and "natural profile" theater. Nobody in the retrieval stack is grading your pie chart.

Counting nofollow as a loss when the page is real. Google has treated nofollow, sponsored, and ugc as hints since 2019, not hard directives (qualify outbound links). Perplexity and ChatGPT can still retrieve a nofollow page. If the prose is the evidence, the rel attribute is a Google hint, not an AI-visibility death sentence.

Keep	Drop	Why the line is there
A dated article that states a fact a buyer would ask	A guest post whose only job is an exact-match href	One is a retrievable source. The other is on Google's spam list.
Your brand name in the sentence, with or without a link	A keyword-stuffed anchor in a footer	Models read the sentence. Google already discounted the footer pattern.
Five witness URLs you can recite	A crawl chart of 500 referring domains	Mentions beat backlink count in the 75k-brand table.
rel="sponsored" on a paid placement	An unmarked paid "editorial"	Policy-legal advertising vs link spam.
The service or research URL that answers the query	A homepage-only link for "brand"	Citations attach to pages that answer, not to billboards.
Search Console used to find junk	Search Console used as an AI score	Hygiene vs a metric the engines do not publish.

What I refuse to put on this table #

I will not give you a guest-post outreach sequence, a HARO clone, or a "10 sites that always link." That is how this spoke turns into a digital-PR clone, and that clone already exists. If the keep column is empty, the work is earning witness pages — not manufacturing hrefs.

I also will not tell you that dropping theater will "not affect rankings." Some of those links were, in fact, holding a classic position. Google still uses the link-or-refer signal. The honest trade: you may give back a little ranking theater to stop feeding a spam pattern, and you should watch Search Console for eight to twelve weeks after you clean a rotten neighborhood. That is a ranking risk you accept on purpose. It is not an AI-visibility loss if those URLs were never going to be retrieved.

---

How do you tell if a third-party URL is working as citation evidence? #

You open the URL, you ask the buyer question in the engines you care about, and you check whether that page — or a page that repeats the same fact — appears as a source. If the only place the href "works" is a crawl tool, it is not working as evidence.

This is a monthly pass, not a new analytics property. Thirty to forty-five minutes. Same ten questions each time. I do not blend ChatGPT, Perplexity, and Google AI Overviews into one "AI backlink score." They are different retrieval stacks.

The four checks #

1. The stranger test. Incognito. No login. Read the linking page until you hit the sentence that names you. Write down the fact in seven words. If you cannot, the page has no extractable claim. Fail.

2. The retrieval test. Ask the same buyer question in Google (note whether an AI Overview appears and which URLs sit on the snapshot), in ChatGPT with search (GPT-5.5 on my box as of September 2026), and in Perplexity. Record three columns: mentioned, cited with a source URL, invisible. A mention without a source is still useful entity corroboration. A source chip on a third-party URL that names you is the evidence win. A source chip on your own URL is a different win — your page got extracted. Do not collapse those two.

3. The metric hygiene test. In Ahrefs or whatever crawl tool you already pay for, ignore the raw backlink total for this pass. Look at whether the linking page is itself a page that ranks or gets cited for anything. If the domain only exists to emit hrefs, it is a neighborhood, not a witness. In Search Console, export latest external links when you suspect spam. You are hunting patterns, not collecting a high score.

4. The consistency test. Does that third-party page use the same legal name, city, and service string as your Organization markup and your homepage H1? If it does not, you did not gain an entity witness. You gained a second, slightly wrong node. Fix the string before you chase another href.

Check	Pass	Fail
Stranger test	You can quote a specific fact	You only find a footer URL or a keyword anchor
Google AI Overview	Your brand or a witness URL is a source chip	You rank under the snapshot with no chip, or no Overview fires
ChatGPT / Perplexity	A source list includes a page that names you	The model names a competitor and cites their witness pages
Crawl-tool count	You can ignore it and still name five witness URLs	The count is the only "win" you can point at
Entity strings	Name, place, and service match your site	The mention is a nickname, an old DBA, or a wrong city

A worked example, without fake clients #

Say the buyer question is "who should I hire for an AI-visibility site rebuild." A useful witness URL looks like a dated industry piece that says a named person rebuilt a site so Google AI Overviews and Perplexity could extract service pages, and it states a constraint — what was in scope, what was not. A useless "backlink" looks like best seo company on a page titled "Submit Your Site — DR 60+ Guaranteed."

I will not invent a before/after citation rate for that example. Seer's April 2026 update is the closest public receipt I trust on the click side of a citation, and they refuse to call it causal. Ahrefs will not convert a 0.664 mention correlation into a promise. Neither will I.

What I log, and what I refuse to log #

I log:

• The ten questions.
• For each engine: mention / citation / miss.
• The witness URL when one appears.
• One follow-up: rewrite our page, or go earn a real witness page. Not both in the same hour if you are a small team.

I refuse to log:

• "AI backlink score."
• A blended SOV that mixes ChatGPT and Google.
• A forecast that turns a new referring domain into next-quarter pipeline.
• Any sentence that starts with "studies show X% of AI citations come from backlinks." I have not seen a primary source that states that, and I will not write one into existence.

If the monthly pass keeps returning misses, the failure is usually one of two things. The witness layer is empty — no third-party URL states a fact. Or your own URLs are unextractable — no question, no first-sentence answer, no entity-clear service page. The second failure is a site problem. That is AIO and AEO work on the public URL, which is the thing I actually build.

What a pass looks like after 90 days #

I do not promise a citation lift at day 90. I promise you will know which object you are missing.

Pattern after three monthly passes	What it usually means	What I do next
Mentions appear, no source chips	Entity corroboration is starting; retrieval is not picking a page	Tighten the witness pages and your own extractable URL for that query
Source chips on third-party URLs, never on yours	The web can prove you exist; your site is still hard to quote	Fix the on-site answer. Do not buy more hrefs.
Source chips on your URL, zero witness pages	Extraction works; the brand node is still self-issued	Earn one real witness URL. Do not celebrate the crawl chart.
Nothing in any engine	You are not in the evidence pile	Stop the link vendor. Write the ten questions. Build or earn pages that answer them.
Crawl-tool count up, same misses	Theater	Fire the volume campaign. Keep the two or three URLs that survived the stranger test.

Ninety days is long enough to see a pattern and short enough that you should not wait for a model retraining myth. Perplexity and ChatGPT with search can fetch a new witness page in days once it is indexed. Google AI Overviews can still ignore you if the snapshot has enough other sources. That is a different clock. Do not average them.

A weekly Google-only citation fraction is a different measurement job. This page stays the evidence-URL filter. I am not going to recast a scorecard here.

---

Frequently Asked Questions #

Do answer engines read the Google link graph the way classic SEO tools do? #

No. Google Search still uses links as one quality signal. ChatGPT and Perplexity retrieve pages and show source URLs. They do not ingest your Ahrefs or Moz graph. Google's own ranking explainer still mentions prominent sites that link or refer to content. Perplexity's API docs describe citations as fetched search_results with real URLs, not as a PageRank export. Treat crawl-tool graphs as a ranking-era map. Treat fetchable pages as the AI-visibility map.

Does domain authority still affect AI citation likelihood? #

As a third-party score, weakly — and only as a noisy proxy. Ahrefs' May 26, 2025 study put Domain Rating at 0.326 against AI Overview brand mentions, behind web mentions at 0.664. The December 12, 2025 rerun said AI Overviews lean on DR a bit more than ChatGPT does, and still called it mid-pack versus branded mentions. I would not buy DR. I would buy a page a stranger can quote.

Do unlinked brand mentions count if there is no href? #

Yes. The mention is often the stronger object for answer engines. Ahrefs found branded web mentions — linked or not — correlated more strongly with AI Overview visibility than backlink counts, and they quote their own content team on unlinked mentions mattering more for GEO than for classic SEO. A link is still better when it is honest, because you get the sentence plus a crawl path. The missing href does not delete the sentence.

Is a nofollow or sponsored link useless for AI visibility? #

Not if the page is real evidence. Google treats nofollow, sponsored, and ugc as hints, and it wants paid placements qualified so they do not pass ranking credit. ChatGPT and Perplexity can still retrieve a nofollow URL. A sponsored article that states a dated fact can be a witness. A followed footer link on a farm page cannot. Qualify the paid ones. Judge the page, not the rel value alone.

Should I still chase homepage links for AI visibility? #

Only when the homepage is the page that answers the question. Most homepages are positioning. Most buyer questions are specific. Citations attach to the URL that holds the extractable answer — a service page, a research note, a comparison, a FAQ. A homepage href can still help classic rank and entity discovery. It is a weak default for AI visibility if that homepage cannot be quoted in two sentences.

Can bought links or expired-domain networks get me cited more? #

They can get you a spam problem. They are a bad bet for citations. Google's link-spam policy names paid links that pass credit and low-value pages built to manufacture linking signals. Expired-domain tricks sit in the same family of "authority that is not yours." Ahrefs' link-volume correlations with AI mentions were the weak ones. I have not seen a dated primary source that says bought neighborhoods raise ChatGPT or Perplexity citations, and I will not invent one.

Do ChatGPT and Perplexity treat backlinks the same way Google Search does? #

No. Google Search still scores a link graph. ChatGPT and Perplexity search the web and cite fetched pages. Ahrefs' December 12, 2025 cut found ChatGPT correlated less with classic authority metrics (DR, backlink count, branded search volume) than Google's AI Mode and AI Overviews. That does not make ChatGPT a loophole for spam. It means a modest brand with real witness pages can show up in chat before it shows up in an Overview. Perplexity will show the URL it fetched. Google may use the href twice — once for rank, once if that URL or yours becomes a supporting link on the snapshot.

How is a backlink different from an AI citation? #

A backlink is an href on someone else's page. An AI citation is your name or your URL appearing as a source inside the answer. You can have thousands of backlinks and zero citations. You can be cited from a third-party evidence URL that never linked to you. Seer's 2026 CTR work is about the citation-versus-uncited gap on Overview queries, with an explicit causality caveat. Ahrefs' 38% top-10 finding is about how often a cited URL also ranked. Neither number is "the percent of citations that come from backlinks," because that study is not on the table.

---

Get a site answer engines can quote — then earn witness URLs, not theater #

A keep-versus-drop list does not mint citations by itself. Google AI Overviews, ChatGPT, and Perplexity still need pages they can lift: question-led headings, first-sentence answers, FAQ blocks, and entity-clear service URLs. If the site is a slogan and a form, no third-party href will save the extraction. If the witness layer is a folder of farm links, no FAQ schema will invent a second domain that names you.

I build Premium AIO/AEO websites for operators who want that extractable layer designed in. The offer is the site — not a referring-domain package, not a guest-post calendar, and not a six-engine "AI backlink" dashboard.

If you want an AI-visibility-ready site built so the evidence you already have has somewhere honest to land, use /contact. Bring the five witness URLs if you have them. If you do not, we write the stranger-test list on the call before anyone talks about links.

SEO-certified since 2021. Founder, AI Systems Architect, Fractional AI CTO. The job this month is not a bigger crawl chart. It is a short list of third-party URLs an answer engine can open — and a site that is still worth citing when they do.