Do backlinks still matter for AI visibility

Yes. Backlinks still matter for AI visibility — as crawl paths, ranking signals, and entity corroboration — not as a switch that buys citations. Google still uses links to find pages and still lists PageRank among core ranking systems. Answer engines cite passages they retrieve and can lift. A referring-domain report is not an AI citation API.

This spoke sits under the Answer Engine Optimization playbook. I have been SEO certified since 2021. That is a checkable credential, not a claimed citation-rate lift from any link package I will invent for your deck.

The short answer

• Keep a real link program for discovery, classic rank, and named mentions. Kill the story that Domain Rating is how ChatGPT or AI Overviews “decide.”
• Ahrefs’ 75,000-brand analysis (May 26, 2025) put branded web mentions at 0.664 Spearman with AI Overview brand visibility, against 0.218 for backlink count. Ahrefs says correlation is not causation. Repeat that hedge every time you quote the table.
• Google’s ranking-systems guide still names link analysis and PageRank as core. Its link best-practices page still says Google uses links to judge relevancy and to find new URLs.
• Google’s May 21, 2025 note on succeeding in AI search lists unique content, page experience, crawl access, preview controls, and matching structured data. It does not list “raise DR.”
• Measure cited / mentioned / absent / hallucinated on a frozen prompt panel. Treat referring domains as an SEO input, not the AEO scoreboard.

Do backlinks still matter after AI Overviews?

They matter for the jobs they always did. They do not replace extractable HTML. If a page cannot be quoted in 60–80 words, more inbound links will not finish the writing.

AI Overviews and AI Mode still sit on Google Search. A supporting link has to be indexed and snippet-eligible in ordinary Search — that is the floor in Google’s AI features documentation. ChatGPT Search and Perplexity quote pages they fetch when search is on. None of those products publish a “pay per citation” link formula.

Job	Still true in 2026?	What a link actually does	What it does not do
Discovery	Yes	Crawlers follow href to find URLs	Guarantee indexation or serving
Classic ranking	Yes, among many signals	Link analysis / PageRank still named as core	Guarantee position 1
Entity / aboutness	Yes	Anchor text and the sentence around it teach the topic	Rewrite a vague homepage
AI Overview supporting link	Indirect at best	Helps the URL exist in the Search index and the candidate pool	Force a citation if the passage is mush
ChatGPT / Perplexity footnote	Indirect at best	Helps the URL get discovered and trusted as a web page	Stamp your brand into the model’s memory
Brand mention without a hyperlink	Often the stronger correlation in Ahrefs’ table	Gives the model language about you	Pass PageRank

Operator rule: if someone sells “AI citation backlinks,” ask which engine, which prompt, and which logged URL. If the deliverable is a referring-domain screenshot, you bought SEO theater.

What job do links still do in Google Search?

Two jobs Google still documents in public: relevancy, and finding pages.

The ranking-systems guide is explicit that ranking is mostly page-level, and that site-wide signals and classifiers also contribute. A handful of good site-wide signals does not make every URL a winner. A handful of poor ones does not sink every URL. Treat “the domain is strong” as a prior, not a citation.

PageRank is not the old toolbar green bar. Google retired the public score years ago. The same guide still says PageRank continues as part of core ranking, and that how it works has evolved. Quote that sentence. Do not quote a vendor “PageRank 2026” number.

Public Google object	What the docs say	Operator translation
Link analysis + PageRank	Core ranking systems; evolved since the original paper	Links still vote. They are not the only vote.
Page-level ranking	Systems are designed to work on the page	A homepage link does not bless a blocked service URL
Site-wide signals	Used; they do not guarantee every URL ranks	Do not skip on-page work because DR looks fine
Crawlable <a href>	Google uses links to find new pages	JS-only clicks and empty anchors fail discovery
Anchor text	Tells people and Google what the target is about	“Click here” teaches nothing
Internal links	Every page you care about should be linked from at least one other page	Orphans do not get cited
Outbound citations	Linking to sources can help establish trustworthiness	Hoarding PageRank by never citing anyone is not a strategy
rel="sponsored" / nofollow	Required when the link is paid for ranking credit	Buying untagged “editorial” links is link spam

Procedure — confirm links are still doing Search work, not just a spreadsheet:

1. Pick ten money URLs. Confirm each has at least one internal href from a crawled page (site search or a crawl export, not a vibe).
2. Inspect five of them in Search Console. Indexable, 200, not accidentally noindex / nosnippet.
3. Sample twenty inbound links. Are they <a href> in article HTML, or widgets, profiles, and footers?
4. Read the sentence around the five you would actually want retrieved. Does it state who you are and what you sell?
5. If the sentence is empty and the URL is a homepage footer, log it as SEO-only. Do not put it on the AEO scoreboard.

Google cannot cite a URL it never found. That is the crawl job. It is necessary and boring.

Why are backlinks not an AI citation API?

A citation is a retrieval event plus a quoting event. The engine has to fetch (or already know) a passage, decide it answers a sub-question, and attribute it. A hyperlink on a random domain is one possible prior. It is not the event.

Google’s succeeding-in-AI-search post is the tell. Unique, non-commodity content. Page experience. Technical access. Preview controls. Structured data that matches visible text. No line that says “acquire referring domains until the Overview appears.”

ChatGPT Search and Perplexity are not Google’s ranker with a new coat of paint. Their source-selection formulas are not public. Do not invent a Bing-PageRank story and call it a citation strategy. Discovery into an index still matters. Quotable HTML still matters more once you are fetchable.

Claim you will hear	What is actually true	What to demand instead
“Links are how AI decides whom to cite”	Unpublished. No vendor has Google’s Overview selector	A logged prompt where your URL is the supporting link
“DR 70 = Overview eligible”	Ahrefs DR is Ahrefs’ metric. Google does not use it	Indexation + snippet eligibility + a quote test
“We will get you cited in ChatGPT this month”	No one can SLA a model’s source panel	Frozen prompts, dates, screenshots
“Guest posts with exact-match anchors = AEO”	That is closer to link spam than to citations	A sentence a stranger could repeat
“Unlinked mentions do nothing”	Ahrefs’ brand-mention correlation was the strongest in that study	Mentions on pages engines already retrieve
“If we rank, we get cited”	Ranking helps the candidate pool. Selection still wants a lift-able span	Answer-first pages before more outreach

Operator rule: APIs have request and response. Link vendors have invoices and referring-domain charts. Those are different products.

How do ChatGPT Search and Perplexity change the link question?

They change the retrieval path, not the physics of hyperlinks. If search is on, the product fetches web pages (or an index of them) and names sources. If search is off, you are arguing with training-data memory. Links do not patch memory. Accurate public sentences on fetchable URLs do more.

ChatGPT Search has historically retrieved against Bing’s index. I am not going to pretend OpenAI published a 2026 PageRank clone, or that I measured Bing’s link graph for you. The operator fact that survives the hedge: a URL that is not in the index the chat product searches cannot be a live citation. Crawlable links still help that discovery job. They still do not stamp a footnote.

Perplexity is similar in kind: it retrieves, then quotes. Fresh pages and community threads show up in a lot of operators’ logs. That is an observation from working panels, not a census. Do not convert it into “Reddit links are the new DR.”

Product	What is public enough to operate on	Link implication	Hedge you must keep
Google AI Overviews / AI Mode	Eligibility = indexed + snippet-eligible Search pages; unique helpful content	Classic links still feed crawl and rank, which feed the candidate pool	Google does not document “DR → Overview”
ChatGPT with Search	Sources appear when search ran; the live fetch path is not your invoice	Be discoverable in the index that product searches; make the passage liftable	Source ranking formula is unpublished
Perplexity	Named sources in the answer UI	Same: fetchable URL + a sentence worth lifting	I will not invent a Perplexity link API
Chat with search off	Memory / prior, not a crawl of your Tuesday post	Links do almost nothing this hour	Do not A/B “citations” against a memory answer

Procedure — separate the products in the log so you stop mixing jobs:

1. For each frozen prompt, record engine and whether live retrieval ran (sources panel vs no sources).
2. If there are no sources, do not file a link ticket. File a “this was memory” ticket.
3. If sources exist and you are absent, list the cited hosts. Those hosts are the media list.
4. If Bing Webmaster Tools (or equivalent) shows the money URL as not indexed, fix discovery before you buy Google-only guest posts.
5. Re-run the same prompt on Google (Overview / AI Mode) and on the chat product. They will disagree. That is normal. Do not average them.

Operator rule: one referring-domain campaign cannot be the strategy for four products. Split the panel or you will “prove” links work on the product they already ranked in.

What did Ahrefs actually measure on mentions versus links?

Louise Linehan’s May 26, 2025 Ahrefs study analyzed 75,000 brands. Method matters, because people paste the 0.664 as if it were a law of physics.

They filtered for domains with Domain Rating above 40 and took each domain’s highest-volume keyword with monthly search volume of at least 800 — a way to find “legitimate brands,” in their words. They used Spearman correlation against AI Overview brand mentions in Ahrefs Brand Radar. About 26% of the brands had zero Overview mentions; they studied the rest.

Ahrefs’ own hedge, on the same page: larger positive values are stronger correlations, and correlation is not causation. All factors they studied sat in the moderate-to-very-weak band on the Spearman scale. Quote that. Do not turn 0.664 into “mentions cause Overviews.”

Factor (Ahrefs, May 2025)	Spearman vs AI Overview brand visibility	What you may not conclude
Branded web mentions	0.664	That buying mention volume causes citations
Branded anchors	0.527	That stuffing brand-name anchors is safe
Branded search volume	0.392	That ads for the brand name mint Overviews
Domain Rating	0.326	That DR is Google’s score
Referring domains	0.295	That more unique linking sites = more citations
Branded organic traffic	0.274	That traffic is the Overview input
Backlink count	0.218	That raw link volume is the lever
Branded ad traffic / ad cost	0.216 / 0.215	That paid search buys Overview mentions

Ahrefs also split web-mention frequency into quartiles. Brands in the top 25% for web mentions averaged 169 AI Overview mentions in their set; the next quartile averaged 14 — the “up to 10X” line on that page. The bottom two quartiles averaged 0–3. That is a description of a skewed distribution in their Brand Radar extract. It is not a promise that your site climbs a quartile if you hire a PR firm.

Two more method traps:

• The study’s outcome is brand mentions inside Overviews, not “your how-to URL was the supporting link for query X.”
• The sample is already-strong domains (DR > 40, fat head brand queries). A five-page local service site is not in that frame.

Treat the table as a reason to stop worshipping backlink count as the AEO KPI. Do not treat it as a reason to zero the link budget. Ahrefs still found positive (weak) correlations for DR, referring domains, and backlinks. Weak is not zero. Weak is not “buy 400 blog networks.”

How operators misuse this study — and the correction:

Misuse	Why it is wrong	What to write in the deck instead
“Mentions are 3× more important than links (0.664 / 0.218)”	Ratios of Spearman coefficients are not effect sizes	“Mentions correlated more strongly than backlink count in Ahrefs’ 2025 brand-Overview extract. Correlation, not causation.”
“We are DR 45, so we are in the study’s world”	The filter was DR > 40 and a fat-head brand keyword ≥ 800 volume	“Our brand query is X volume. If it is tiny, this table is a cousin, not a mirror.”
“Buy mention volume until we hit the top quartile”	Quartiles describe a skewed snapshot, not a treatment effect	“We will earn sentences on hosts already in our prompt log.”
“Unlinked mentions replace SEO”	Backlinks still correlate weakly-positively; crawl and rank still use links	“Keep a real link program. Stop using count as the citation KPI.”
“This proves Overviews ignore PageRank”	The study did not measure Google’s internal PageRank	“Google still documents PageRank. Ahrefs measured brand mentions vs Ahrefs’ own link metrics.”

If you cannot put the hedge on the same slide as the 0.664, do not put the 0.664 on the slide.

How do AI Overviews actually pick supporting links?

Google does not publish the selector. It publishes eligibility and quality floors. Work those. Guessing “they use our Moz DA” is fan fiction.

From the AI-features docs and the May 2025 Search Relations post, the operator stack is still Search:

Floor	Pass	Fail	First fix
Technical access	Googlebot allowed, HTTP 200, indexable HTML	Blocked, noindex, soft 404	robots / Inspection
Snippet eligibility	No accidental nosnippet on the answer	Preview controls strip the passage	Remove the directive
Unique content	A fact or table that is not a clone	Commodity rewrite of the SERP	Original span on the URL
Visible = structured	schema matches the page	Fake FAQ in JSON-LD only	Make the HTML match
People-first	Answers the question a human typed	Written for a crawler costume	Quote test

Procedure — test the page before you buy another link:

1. Copy the first 80 words into a blank note. If a stranger cannot repeat the answer, stop outreach.
2. Confirm the URL is indexed and snippet-eligible.
3. Run the prompt that should cite it. Log cited, mentioned without a link, absent, or hallucinated.
4. If you rank in the top 10 and still lose the Overview, the next ticket is extractability or fan-out, not DR. That split is the point of the ranked-but-missing problem: links already did enough to rank.
5. If you sit at 12–20 and falling, rank recovery — including earned links — is back on the table.

Supporting links in Overviews are still web pages. They are not a parallel index you enter by purchasing “AEO backlinks.”

Rank versus citation is the split most decks flatten:

You observe	Likely job	Link spend	On-page spend
Not indexed	Discovery	Internal href + a few real externals	Make the HTML exist
Indexed, position 15–30	Classic rank	Maybe, if competitors’ editorial links are the gap	Keep the quote test anyway
Position 3–8, no Overview citation	Selection	Usually postpone	Lead, table, fan-out H2s, fact match
Cited on brand prompts, absent on category prompts	Coverage	Mentions on category hosts already in the log	Cluster spokes, not more homepage links
Cited, but the answer invents your price	Trust / conflict	Stop amplifying the wrong sentence	Fix every URL that states the offer

Googlebot following a link and an Overview quoting a passage are different events that happen to share a web. Do not staff them as one ticket.

When is more link work still rational?

When the URL is not even in the candidate pool, or when the engines keep citing a third-party page you do not exist on. Links are a tool. Use them on the actual gap.

Situation	Link work this month?	Why	Pair it with
Money URLs not indexed; few crawl paths	Yes — internal first, then a few real externals	Discovery is the bottleneck	Crawlable href, XML sitemap, Inspection
Rank 12–20 and dropping on the money queries	Yes, if links are the gap vs competitors	Classic rank still feeds Search	On-page that already passes a quote test
Rank 3–8, Overview cites a competitor how-to	Usually no as the first spend	You are already relevant enough to rank	Answer-first lead + table
Overview cites docs, .edu, or a standard you are absent from	Yes — earn a mention on those hosts	Retrieval already likes those URLs	Locked fact sheet so they do not invent your offer
Prompt panel hallucinates your category	Maybe — accurate third-party sentences	The model needs language that is not yours	Correct the homepage first or the clip will disagree
Prompt panel never names you, and you have almost no web mentions	Yes — digital PR aimed at retrieved domains	Ahrefs’ strongest correlation was mentions	Not a 500-site guest-post blast
You already have plenty of homepage footers	No	Those rarely get retrieved	One sentence on a page the engine already cites

Checklist before you open an outreach tool:

• Quote test passed on the URL you want cited
• URL is indexed and snippet-eligible
• You can name the gap: crawl, rank, missing third-party sentence, or hallucination
• Target domains already appear in your citation log, or they are primary sources the log already prefers
• The pitch includes a one-sentence offer a journalist can paste without lying
• You will re-run the same prompts 4–6 weeks after the URL goes live

If you cannot tick those boxes, you are shopping for links because the dashboard looks empty. That is a reporting problem.

When should you postpone backlink outreach?

When the page would embarrass you if it were the supporting link. More inbound votes for a mushy lead just scale the mush.

I have shipped hundreds of production sites. The pattern that wastes a quarter is a link sprint on a homepage that still will not say what the company sells. The Overview does not owe you a footnote for effort.

Postpone if	What you are actually seeing	Do this first
noindex / blocked money URLs	Nothing to retrieve	Technical eligibility
nosnippet on the template	Google cannot use the passage	Preview controls
First 80 words are brand poetry	Unquotable	Answer-first rewrite
About, schema, and footer disagree on city or legal name	Entity conflict	Fact sheet
No frozen prompt list	You will argue from vibes	20 prompts, one sheet
Rank is fine; citation is missing	Selection / extractability	Tables, definitions, fan-out H2s
Legal rebrand in flight	Any clip will go stale	Wait until the name is stable
Vendor wants exact-match guest posts	Spam risk	Walk away

Checklist for a one-week freeze on outreach:

• Five money URLs pass the 80-word quote test
• Internal links from the cluster hub actually exist as href
• Search Console: indexed, not excluded, not snippet-stripped
• One entity fact sheet the PR person is not allowed to “improve”
• Prompt panel logged once, dated

Then you may earn links. Not before. Bravery is not a restore strategy, and neither is a referring-domain target pulled from a competitor’s Ahrefs chart.

Which placements help crawl, entity, and retrieval?

Sort every placement by whether a retrieval system would ever fetch that URL for your buyer’s question. A follow link on a page nobody asks about is SEO residue.

Placement type	Crawl / rank	Entity language	Retrieval for answers	Default call
Editorial mention + descriptive anchor on a page already in your citation log	Useful	Strong if the sentence is accurate	Best case	Do this
Unlinked brand mention on that same class of page	Little PageRank	Strong in Ahrefs’ mention correlation	Often enough	Still do this
Primary-source cite (docs, standard, dataset) pointing at your method URL	Useful	Strong if they use your name correctly	High	Pitch the method, not the homepage
Homepage footer on a DR-looking roundup	Possible drip	Weak (“visit site”)	Rarely retrieved	SEO-only; do not report as AEO
Guest post with stuffed commercial anchors	Spam-shaped	Looks like advertising	Low, plus policy risk	Skip
Profile / directory / bookmark farms	Noise	Noise	Almost never	Skip
Paid advertorial without rel="sponsored"	Policy violation	Contaminated	Not worth it	Qualify or do not buy
Internal contextual links inside a cluster	Discovery on your site	Aboutness for your own URLs	Helps Google find the spoke	Required, not optional
“Click here” in a press release blast	Weak	Weak	Weak	Rewrite the sentence or skip

Procedure — score a prospect URL in five minutes:

1. Search your frozen prompts. Does this host already appear as a cited source? Yes beats a higher DR host that never appears.
2. Open the live article, not the publisher’s homepage. Would a model fetching this URL learn a correct fact about you?
3. Check the intended sentence. Brand name, category, constraint. No adjectives you cannot defend.
4. Check the link: real <a href> to the URL you want retrieved (often a spoke or method page, not /).
5. If it is paid, it gets sponsored or nofollow. If the seller refuses, you are buying link spam.

A mid-tier trade URL that already shows up in Perplexity for your category beats a national lifestyle footer. Retrieval is the filter. Vanity is not.

How do internal links and clusters interact with off-site links?

Off-site links cannot repair an orphan spoke. Google’s own link-best-practices page says every page you care about should be linked from at least one other page on your site, and that internal anchor text helps both people and Google.

Content clusters for AI visibility are how you point that equity at a question family instead of spraying it at a bloated blog index. The pillar answers the hub question. The spoke answers one question all the way down. Internal href with honest anchors is the wiring.

Internal move	What it helps	What it does not replace
Pillar → spoke with the question as anchor	Discovery + aboutness	A third-party corroborating sentence
Spoke → pillar with the offer as anchor	Hub strength	Indexation if the spoke is noindex
Spoke → sibling for the next sub-question	Fan-out coverage	A unique table you never wrote
Footer sitewide keyword stuffing	Spam-shaped	Anything
Orphan “AEO” URL created for a vendor	Confusion	Crawl

Procedure — one afternoon on internal links, before another outreach sequence:

1. List the ten URLs you actually want cited. If a URL is not on that list, it does not get a campaign.
2. From the cluster hub, add or fix contextual links with anchors that match the question the spoke answers.
3. From each spoke, link the hub and one sibling. No chain of ten “related posts” widgets as a substitute.
4. Recrawl in Search Console or fetch as Google. Confirm the href exists in rendered HTML.
5. Only then decide whether an external mention is still the missing layer.

Operator rule: internal links are the cheapest PageRank and aboutness work you will do this quarter. Do them. Then buy fewer, better externals.

What fails when teams buy Domain Rating packages?

The failure mode is a paid “50 DR 40+ posts” blast while the money URLs stay unquotable. Rank barely moves. The prompt panel does not move. You now have a cleanup problem.

Google’s spam policies define link spam as creating links primarily to manipulate rankings: buying or selling ranking credit, guest posts with optimized anchors, low-quality directories, sitewide footer schemes, automated link vendors. Paid links are allowed when they are qualified with rel="sponsored" or nofollow. Unqualified “editorial” inventory is the product being sold. It is also the product that gets you into a spam classifier.

What you bought	What breaks	What it costs	What you do instead
Guest-post network with commercial anchors	Possible link-spam action; nothing retrieved for answers	Cleanup, disavow theater, delayed real pages	Stop. Qualify remaining paid links. Rewrite money URLs
Homepage links on unrelated roundups	Spreadsheet looks busy; citation log does not	Three months of “we are building authority”	One retrieved-host mention with a correct sentence
Exact-match anchors at scale	Looks like 2012	Rank risk	Brand-name or descriptive anchors, sparsely
“AI citation backlink” SKU	The team waits on a fake API	Opportunity cost vs extractability	Quote test + eligibility + prompt log
Directory / profile packs	Index bloat, association with junk	Time	Delete from the report

Concrete cost, without inventing a client: you spend the quarter on inventory Google’s policies already describe as manipulative, while a competitor ships a dated table and gets the supporting link. You then spend the next quarter arguing about disavow files. The Overview still quotes their table.

I will not attach a fake recovery percentage to that story. The recoverable move is operational: pause the vendor, screenshot the live guest posts, qualify or remove, and put the next dollar on HTML that can be lifted.

Cleanup procedure when the package already shipped:

1. Export every placement URL the vendor delivered. Open them. Screenshot date, anchor, rel, and the sentence.
2. Mark each row: real editorial, paid-unqualified, directory, dead, or redirected to a casino. Be honest.
3. Qualify anything that was paid and still live (sponsored or nofollow) if you control the relationship. If you do not control it, document it.
4. Stop reporting those domains as “AEO wins.”
5. Do not start a dramatic public disavow campaign as theater. Disavow is a last-resort spam tool, not a citation strategy. Use it if Google’s own spam situation applies, not because a dashboard looks dirty.
6. Put the next two weeks on quote tests and internal links. Outreach resumes only against the retrieved-host list.

Checklist that you are actually out of the hole:

• Vendor paused in writing
• Placement inventory exists as a sheet, not a memory
• Paid rows qualified or logged as out of your control
• Prompt panel re-run once so you have a post-junk baseline
• Next dollar assigned to a named money URL, not to “more DR”

How do you measure whether backlink work is helping AI visibility?

Do not use referring-domain count as the AEO KPI. Use it as an SEO input. The AEO scoreboard is the prompt panel plus eligibility.

Freeze 20–40 prompts that a buyer would actually type. Re-run on a cadence. Score each row. Change one cluster of work at a time or you will never know what moved.

Metric	What it tells you	What it hides	Cadence
Cited (your URL in sources / supporting links)	Retrieval + quoting	Quality of the visit	Weekly, same prompts
Mentioned, no link	Language exists; credit is partial	Whether the fact is accurate	Same
Absent	You are not in the answer	Whether you rank in classic Search	Same
Hallucinated	Dangerous; often a missing or conflicting fact	Link equity	Same, plus incident log
Indexed / snippet-eligible	Floor	Selection	After each ship
Rank on the money query	Classic candidate pool	Overview selection	Weekly GSC
Referring domains / DR	SEO prior	Citations	Monthly, not weekly panic
New URLs on hosts already in the citation log	AEO-aware PR	Vanity DR	Per placement

Checklist for a measurement sheet that survives an exec meeting:

• Prompt text frozen (no “we tweaked the wording”)
• Engine + date + cited domains pasted
• Your URL: cited / mentioned / absent / hallucinated
• Fact-match against the entity sheet (name, offer, city, price band)
• Search Console: query, position, index state
• Placement URL if you earned something that week (live href, rel, sentence)
• One change log line (“shipped quote-test on /services”, not “did SEO”)

If referring domains rose and the panel did not move, the link work did SEO (maybe) and did not do AEO (yet). Report both. Do not average them into a “visibility score.”

Ahrefs Brand Radar can track branded Overview mentions if you already pay for it. It is still a brand-mention product, not a URL-level citation log. Use it as a complement. Do not replace the screenshots.

A single prompt row should look like this — copy the columns, not my fictional URLs:

Field	Example of a real row	Fail
Date	2026-08-22	“sometime this month”
Prompt	“Do backlinks still matter for AI visibility?”	A paraphrased cousin each week
Engine	Google AI Overview · ChatGPT Search on · Perplexity	“AI” as one bucket
Your status	cited / mentioned / absent / hallucinated	A 1–100 “visibility score”
Cited hosts	list the actual domains	“lots of big sites”
Your URL if cited	full URL	Homepage assumed
Fact-match	offer sentence correct / wrong city / invented package	Skipped
Rank (GSC)	query + position that week	Sitewide average position
Placement shipped	live article URL + sentence, or “none”	“outreach in progress”

If you cannot fill those columns, you are not measuring AI visibility. You are decorating a backlink report.

What should a week versus a quarter look like?

A week can stop you from buying junk and can make the money URLs fetchable and quotable. A week cannot manufacture a mention graph. A quarter can earn a handful of retrieved-host sentences if the HTML is already honest.

One-week procedure:

1. Freeze 20 prompts. Run them once. Save the panel.
2. Quote-test five money URLs. Rewrite the leads that fail. Use answer-first as the craft, not a new file format.
3. Fix internal href from the hub to those five. Kill accidental nosnippet / noindex.
4. Inspect in Search Console.
5. Build a target list of 15 hosts that already appear in the panel. Not a DR leaderboard.
6. Write a one-sentence offer and a two-sentence method blurb. Lock them. Outreach uses those sentences or it waits.
7. Decline any vendor whose deliverable is “X posts, DR Y, guaranteed citations.”

90-day procedure:

1. Keep the same prompts. Re-run at day 14, 45, and 90.
2. Earn or pitch 3–8 placements on the retrieved-host list. Prefer a correct unlinked mention on a fetched page over a follow link on a ghost URL.
3. If rank on money queries is still the hole, add classic digital PR and real editorial links. Still qualify paid ones.
4. Ship one original extractable object (table, numbered method, dated number with n) on the URL you want as the supporting link.
5. Do not expand the cluster until those URLs agree with About and schema.
6. Report the panel. Report indexation. Report placements as URLs and sentences. Leave DR in an appendix.

Horizon	In scope	Out of scope
Week	Eligibility, quote test, internal links, frozen log, kill spam vendors	Wikipedia campaigns, 100-site blasts, “citation SLA”
30 days	First retrieved-host mentions, one original table	Rebuilding the whole blog
90 days	A few durable placements + rank recovery if needed	Guaranteed Overview share
Ongoing	Cadence on the same prompts	New metrics every sprint

When this is not worth doing yet: the site is not snippet-eligible, the company cannot agree on a one-sentence offer, or nobody will keep a prompt log. Link outreach on top of that is spend. Fix fetch and honesty first.

A visibility audit is the right hammer when the referring-domain chart and the citation log disagree and the argument has gone political. DIY the week-one table if you have Search Console and a writer who will put a real sentence on the page.

Do not run these in parallel or you will not know what moved:

Parallel mess	Why it blinds you	Sequence
Guest-post blast + homepage rewrite + new cluster	Any panel change is unclaimed	Quote test, then internal links, then a few externals
New llms.txt + DR package	Neither is the selector	Skip both as AEO tactics; fix HTML
Changing prompts every week + changing pages	You moved the exam	Freeze prompts for 90 days
Three vendors, one scoreboard	Everyone claims the same citation	One owner of the panel

Lane map for the work itself still lives on /visibility. The audit is for when the spreadsheet and the screenshots cannot be in the same meeting without a referee.

FAQ

Do backlinks still matter for AI visibility?

Yes. They still help Google find pages, they still feed classic ranking through link analysis and PageRank, and they still carry entity language when the sentence around the link is accurate. They are not an API that forces AI Overviews, ChatGPT, or Perplexity to cite you. Measure citations on a frozen prompt panel; treat links as infrastructure.

How do I measure whether backlink work is helping AI visibility?

Freeze 20–40 prompts and log cited, mentioned, absent, and hallucinated, plus fact-match, on a fixed cadence. Watch indexation and classic rank as the candidate-pool inputs. Put referring domains in an SEO column, not in the AEO total. If links rose and the panel did not move, say that out loud.

What usually fails first when teams try this?

They buy DR packages or guest-post networks while the money URLs still fail an 80-word quote test. The second failure is targeting high-DR homepages that never appear in the citation log, so nothing gets retrieved. The third is reporting Domain Rating as if it were Overview share.

How long does this take to show results?

Eligibility and internal-link fixes can show up on the next crawl — days to a few weeks, not a promise. Earned mentions on retrieved hosts are slower; plan a 4–6 week retest after the live URL. There is no citation SLA, and I will not invent a lift curve for your deck.

What should I skip if I only have a week?

Skip guest-post blasts, directory packs, exact-match anchor campaigns, and any vendor selling guaranteed AI citations. Spend the week on a frozen prompt log, quote-tests on five money URLs, internal href from the hub, snippet eligibility, and a 15-host list drawn from the log.

When is this not worth doing yet?

When money URLs are blocked or noindex, when the offer sentence is not agreed, or when nobody will keep a prompt panel. Link building cannot cite a page Google cannot fetch, and it cannot correct a category you will not state. Fix eligibility and honesty first.

CTA

If your link report is up and the prompt log still omits you, stop buying DR — make the pages quotable, then earn mentions on hosts the engines already retrieve.

Lane: /visibility · Book a visibility audit.