Retrieval-shaped content
Your highest-intent pages restructured so each section opens with a self-contained answer a model can lift without rewriting. This is the fastest-moving lever and usually the first thing we ship.
How a model decides who to cite
Generative engines answer from two places: a live retrieval pass over the web, and what the model absorbed in training. Retrieval favours content that is crawlable, extractable and recently updated. Training-era memory favours brands mentioned repeatedly across many independent sources over a long period.
That split explains why citation work feels slower than SEO. You can influence retrieval in weeks by restructuring pages. You cannot influence what a model already believes about your category in weeks — that is accumulated third-party mention volume, and it moves at the speed of PR and community presence.
It also explains why the winners in AI answers are frequently not the companies with the best websites. Models lean heavily on comparison posts, review platforms, forum threads and documentation — sources you do not own. A brand with mediocre SEO but heavy Reddit and G2 presence often outranks a polished site nobody discusses.
So the work runs on two tracks at once. The fast track makes your own content retrievable: answer-first structure, clean extractable passages, entity markup, a sane crawler policy. The slow track builds presence in the places models already trust, which means earning mentions rather than publishing more pages.
Neither track is guesswork, because both are measured against the same panel. If a change moves the citation rate, we do more of it. If three months of a tactic moves nothing, we say so and stop — which is the part of this discipline most agencies skip, because it requires admitting something did not work.
What the engagement delivers
The engagement delivers restructured content models can lift, an earned-mention programme in the sources they retrieve from, original data worth citing, and a weekly panel re-run showing whether any of it moved your share of voice. Everything is reported against the baseline.
Third-party mention programme
Placements in the comparison posts, review platforms and documentation that your panel shows models actually retrieve. Earned, not paid, and targeted at sources rather than domain-authority scores.
Original data assets
Research a model has a reason to quote, because nobody else published the number. This is the most durable citation source and the slowest to build — expect one substantive asset per quarter.
Entity and markup work
Consistent naming, structured data and knowledge-graph presence so engines treat your brand as one resolvable entity rather than a string that might mean several things.
Weekly panel report
The same two hundred prompts, re-run, scored and charted against the baseline. One page, sent weekly, showing movement or the absence of it.
Quarterly rescope
What worked, what did not, and what we are stopping. Tactics that fail to move the panel in a quarter get dropped rather than defended.
How the programme runs
The programme runs in monthly cycles against a fixed baseline: shipping retrieval fixes first because they move fastest, then building earned mentions and original data, with the panel re-run weekly throughout so every tactic is judged on evidence rather than opinion.
- 01
Baseline and target prompts
We start from a GEO audit, or run one. Without a before number there is no way to prove a change worked, and this discipline attracts enough hand-waving already.
→ Scored baseline and the prompts we are targeting
- 02
Retrieval fixes shipped first
Answer-first restructuring, extractable passages, schema and crawler policy on the pages closest to breaking through. These land in weeks and often move the panel before anything else has started.
→ Restructured pages, with before-and-after scores
- 03
Earned presence in trusted sources
We go after the specific sources your panel showed models retrieving — the comparison post that keeps getting cited, the forum thread, the review platform — rather than a generic link campaign.
→ Placements in sources the panel identified
- 04
Original data worth quoting
One research asset per quarter that produces a number nobody else has. Slow, expensive, and the only lever that keeps working after the technical wins flatten.
→ A published, citable data asset
- 05
Weekly measurement and quarterly rescope
The panel re-runs weekly. Every quarter we show what moved, name what did not, and change the plan accordingly rather than repeating last quarter with more volume.
→ Weekly trend and a rescoped next quarter
Which levers move citations, and how fast
The levers differ sharply in speed and durability. Restructuring your own content is fast and cheap but capped. Earning third-party mentions is slow and compounding. Original data is slowest of all and the only lever that keeps producing citations long after you stop working on it.
| Lever | Time to move | Durability |
|---|---|---|
| Answer-first restructuring | 2–6 weeks | Capped — works until competitors do the same |
| Schema and entity signals | 3–8 weeks | Durable, but a hygiene factor rather than an edge |
| Crawler policy and llms.txt | Days | Binary — it either blocks retrieval or it does not |
| Comparison and alternatives content | 1–3 months | Strong, because models retrieve this shape heavily |
| Third-party mentions | 3–6 months | Compounding, and hard for competitors to undo |
| Original research and data | 4–9 months | Longest-lived — a quoted number keeps being quoted |
Signals you need this now
You need citation work when a baseline shows models naming competitors instead of you, when your category is being summarised by assistants before buyers reach any website, or when strong classic rankings are no longer converting into pipeline the way they used to.
- A GEO baseline showed you absent or barely mentioned
- Competitors are the default answer for your category's buying questions
- Rankings hold steady but organic pipeline is falling
- Your buyers research through assistants before they visit vendor sites
- You publish good content that models never seem to retrieve
- You have original data nobody has packaged for citation
- Third-party coverage of your brand is thin or outdated
What movement looks like
Movement is uneven and worth setting expectations about. Retrieval fixes often show within six weeks. Mention-driven gains appear from month three. Some prompts never break through because an entrenched competitor owns the source set, and we identify those early rather than spending against them.
Where this fits
Related GEO services
Common in these sectors
The adjacent disciplines
Deciding between options
Questions about LLM citation acquisition
Each answer is written to stand alone in 40 to 60 words — the shape an AI Overview or Perplexity citation lifts. Ships with FAQPage schema.
Can you guarantee ChatGPT will cite us?
No, and any agency promising that is selling something. We commit to a measured baseline, a defined scope and weekly reporting against agreed indicators. If those indicators do not move within the agreed window, we rescope at no extra cost rather than continuing to bill.
How long before citations start appearing?
Retrieval-driven gains often show within six weeks, because restructuring content is a fast lever. Mention-driven and training-corpus gains typically take three to six months. Categories with an entrenched incumbent take longer, and we flag those before you commit rather than after.
Is this just link building with a new name?
No. Link building optimises for authority metrics; citation work optimises for whether a model retrieves a specific source when answering a specific question. We target the exact pages your panel shows engines pulling from, which are frequently forums and comparison posts rather than link-worthy publications.
Do we need the baseline audit first?
Effectively yes. Without a before number there is no way to prove any of this worked, and the audit also tells us which prompts are winnable. If you have a credible baseline from elsewhere we will work from it rather than charging you twice.
What if our category has an entrenched leader?
We will find that in the baseline and tell you plainly. The realistic play is usually to target the specific prompts where the incumbent is weak — narrow, high-intent, comparison-shaped questions — rather than contesting the broad category term you will not win.
Who does the writing and outreach?
We do, unless you prefer your own team executes from our specifications — roughly half of clients choose that. Original data assets always require about thirty minutes of a subject expert from your side, because the credibility comes from your knowledge, not our drafting.
Find out whether this is your constraint.
Thirty minutes with a senior strategist. We pull your live visibility while we talk and tell you plainly whether a LLM citation acquisition is what you need — or whether your problem sits somewhere else.