AI Citation Optimization
A citation is a sentence a model can attach to a URL it can fetch. We write that sentence. We check the live page actually states it. We sample whether assistants then attach your URL, a rival URL, or nothing. We do not sell a footnote. We do not write ChatGPT copy as the product. We do not screenshot a Knowledge Panel.
Position
Assistants invent footnotes. Some attach a URL. Some name a firm and attach nothing. Some attach a competitor because that page states the fact in one clean paragraph and yours hides it in a chart. This page owns eligibility: the sentence, the source that can carry it, and the sample that shows whether attribution happened. Search intent on that source still has to match a question a buyer would ask. A slogan is not a claim.
Perplexity visibility owns how that product shows sources in its own interface. A cite object any model might attach stays here. Open that page for the Perplexity surface. Open this page when the claim itself is not yet worth attributing.
Open Perplexity visibilityDigital PR owns pitches and articles you do not yet have. An existing piece can corroborate a claim if it already states it. We will not run outreach from a cite ticket.
Open Digital PRKnowledge Graph optimization documents facts Google already shows in a Knowledge Panel. A cite is not a panel. A panel screenshot is not a source URL for a claim.
Open Knowledge Graph optimizationBrand mention optimization finds unlinked or wrongly attached name drops. If the assistant says your name and cites nothing, start there. If it attributes a claim to a URL, the object is this page.
Open brand mention optimizationBrand mention work finds names with no source. LLM brand visibility checks the string models speak. Knowledge Graph work starts from a public panel. Digital PR earns coverage you do not have yet. Work this URL if you can name a sentence and a page that should prove it. A newsroom, a panel box, or ChatGPT-style rewriting of the answer belongs elsewhere.
We will not promise Perplexity will cite you. We will not treat one assistant UI as the whole program. We will not sell a citation quota. If the claim is true only in a sales deck, we name that first. If two pages on your site contradict the number, we name that second. Eligibility is the job. A purchased footnote is not.
When this fits
This page fits when you can name a sentence you want attributed and a URL that should carry it. If you needed a newsroom, a panel pass, or a hunt for unlinked names, start on those URLs.
The team recites a statistic. The live HTML does not. We write the claim as a sentence and say whether any fetchable URL currently carries it. Keyword targeting that statistic in a title tag is not a cite object.
A model that attributes anything will pick one. We list the colliding sentences. We name which URL is allowed to carry the claim. Topical relevance of a blog post is not a reason to keep two numbers.
The fact may be yours. The attached URL is not. We sample answers, record the attached sources, and say whether your page is eligible yet. Semantic keywords on the About page will not steal a cite from a page that never states the claim.
We will say so. Eligibility work can still make a sentence fetchable. A purchased footnote is not a deliverable. If the unknown is how models speak the brand, that is LLM brand visibility.
Process
Four jobs, in that order. Each job produces a dated note. We do not open a coverage campaign in the last job. The point is a cite object, not a footnote a vendor promised.
We write the claim you want attributed, in words a human would also accept. Search intent has to match a question a buyer would actually ask. A feature list is not a claim. A vague superlative is not a claim. We date the sentence. We name who will stand behind it.
Outcome A dated claim list, including which sentences we refuse to make citable.
We open the URL that should carry each claim. We check published HTML, not a CMS preview. If the sentence is in an image, a canvas, or behind a form, the source fails. Keyword density is not a source score. Nearby topic coverage is not a source.
Outcome A fetch note: which URL states the claim, which URL contradicts it, which URL cannot be fetched.
We sample assistant answers for the questions tied to those claims. We record whether a URL was attached, which URL, and whether the attached text matches the claim. Entity names in the answer are a side note here. The object is the attribution, not the spoken brand string.
Outcome An attach sample: question, answer date, attached URL or none.
We set a recheck on the same questions after the source page changes. LLM visibility as a name line stays on the brand page. Mention hunts stay on mentions. This retest is whether attribution moved, stayed on a rival, or still has no URL.
Outcome A retest date and the same question list, not a citation quota.
Deliverables
You leave with a claim list, a fetch note, an attach sample, and a retest date. You do not leave with a purchased footnote, a ChatGPT script, or a ranking date.
Sentences you will stand behind, dated, with a refuse pile for slogans and unsourced superlatives. Last quarter's stats deck is not this list.
Which live URL carries each kept claim in fetchable HTML. Colliding numbers are named. Unfetchable placements are named.
Questions, dates, and attached URLs or the absence of a URL. A screenshot of one lucky footnote is not a sample.
Image-only claims, login walls, contradictory pages, and third-party sources that currently win the attach. Each gap has an owner.
When to rerun the same questions. A weekly citation score is not a retest. A promise that Overviews will pick you is not a retest.
Benefits
The claim is no longer trapped in a chart. A retriever can lift it.
Two URLs stop teaching two facts. The fetch note names the keeper.
You see which directory or reseller currently carries the claim. You stop guessing.
Someone reruns the sample. Nobody pretends a plugin bought a cite.
Methodology
This works when someone can change the live HTML of the source URL, and someone can stand behind the number. If both are missing, we are collecting screenshots of other people's cites. We are not making a claim eligible.
We start with the claim in writing, the candidate URL, and a person who can edit that URL. If any of those is missing, we name it first.
We fetch published HTML. We do not trust a Figma frame or a logged-in preview. Does this paragraph match the question a buyer would ask? Is another of your URLs teaching a different number?
You may rename the roles below. You may not leave the claim implied. An implied statistic is how a model cites a reseller.
First working session
The first hour is one claim and one URL. It is not a tour of citation dashboards.
We write the sentence. We open the URL that should carry it. If the sentence is not in the HTML, that is the job. If the unknown is unlinked name drops, that is brand mention optimization. If the unknown is how the brand is spoken, that is LLM brand visibility. If the unknown is new coverage, that is Digital PR. If the unknown is one product's source UI, that is Perplexity visibility.
Then we fetch the page and sample a small set of questions. We record attached URLs. The extra or missing attaches are usually the job. A generative-engine inclusion program is another child. This hour does not become that program.
We leave with a first claim list or a stop. A list names the sentence, the URL, and who can edit it. A stop names what is missing: a claim legal will sign, a fetchable page, or a job that is actually a mention hunt. Both are outcomes. A citation shopping list is not.
AI search
A fetchable claim can help later extraction. This page cannot sell AI Overviews, a chat mention, or LLM visibility as a score.
People ask whether a cleaner paragraph will land the brand in Google AI Overviews (AIO). If the claim is unfetchable, generative search has nothing honest to attach. That is ours to tidy as eligibility. If the page already states the claim and models still attach a rival, that is still an attach sample on this page, not a purchased Overview. Measuring how the name is spoken stays on LLM brand visibility. Binding two companies in the model stays on AI entity optimization.
AI search needs a page it can fetch. A sentence that only exists in an image does not help. We do not sell Overview presence.
We check whether a URL was attached. That is the cite object. How the assistant pronounces the brand is another URL.
Generative engine optimization owns retrieval and inclusion tactics. This page owns whether the claim is eligible to be attributed. Both stay live.
Schema
JSON-LD helps a machine find a fact the page already states. This work can flag a claim that only exists in markup. It is not a schema campaign as the whole job.
If schema asserts a number the paragraph never says, extractors guess. We name the mismatch. We do not stuff extra claims into JSON-LD to chase a footnote.
We refuse fake Review or FAQ blocks that exist only to host more numbers. A real claim can live in the body.
Markup will not force generative search to cite the URL. Eligibility is not a purchased mention.
Who we work with
Someone has to put the sentence on the page. We work with that person. We do not replace them with a second citation retainer.
Signs the sentence. Holds the refuse pile when marketing wants a superlative in the sample.
Prints the claim in HTML. Rechecks after a template change that might hide the paragraph.
Can keep the question list. If they try to run this as unlinked mention cleanup, two jobs collide. Mentions stay on brand mention optimization.
Questions
Eligibility means a model can attach a specific sentence to a URL a crawler can fetch. The sentence must live in HTML, not in a chart or a locked file. We check that evidence. We do not sell the attach. A brand name with no URL is a mention, not a cite.
No. Models choose what to attribute. We make the claim worth attributing: one sentence, one live URL that states it, and a later sample of whether that URL was attached. We will not sell ChatGPT prose, a Perplexity-only pass, or a place in Google AI Overviews (AIO).
No. Knowledge Graph optimization documents facts Google already shows in a Knowledge Panel. This page never starts from a panel screenshot. It starts from a sentence you want attributed and a URL that should carry it.
No. That is brand mention optimization. This page needs a claim that could be a footnote. Mentions log the name with no source or the wrong source. Do not mix the two tickets.
Then it is not eligible yet. We write that finding. Eligibility begins when published HTML states the claim in text. A hero image, a canvas, or a login wall is evidence of a gap, not a cite object.
Evidence before the attach
Paste the source sentence in plain text, the URL that should carry it, and two assistant answers if you have them. We will say if the claim is unfetchable, if two of your pages disagree, or if the real job is a mention, a spoken name, or a panel fact.
Sentence, fetch, attach, retest. Not a citation shop.
