For most of the last decade, buying search visibility meant buying rankings. That equation has quietly stopped working. When an assistant answers a question, it names a handful of sources — and everyone outside that handful is invisible, regardless of where they sit in the blue links.
The scale of the decoupling is now measurable. Ahrefs’ 2025 analysis of 1.9 million AI Overview citations found that 76% of cited pages also ranked in Google’s top ten. Repeating the study in early 2026 across roughly 4 million cited URLs, that share had fallen to 37.9%, with the remainder split almost evenly between pages ranking 11–100 and pages not ranking at all. The methodologies aren’t identical, so the drop overstates the change — but the direction is not in dispute. This is why hiring an ai search optimisation agency has become a distinct procurement decision rather than a line item inside an existing SEO retainer.
What the data should change about your brief?
The second uncomfortable finding is that the surfaces barely agree with each other. Large-scale citation indexes published through 2026 put domain overlap between ChatGPT and Perplexity at roughly 11%. Google’s own AI Overviews and AI Mode cite the same URLs only about 14% of the time. A brand can be the default answer in one assistant and entirely absent from another.
Two practical consequences follow. First, any agency selling a single “AI visibility score” is compressing four or five uncorrelated channels into one number that can’t be acted on. Second, because so many citations come from pages with no organic footprint, the work skews toward things classical SEO teams often treat as secondary — entity consistency, third-party mentions, structured data, content freshness and being quotable in a literal sense.
Five questions that separate operators from opportunists
| Ask this | A workable answer | Warning sign |
| How do you measure citations? | Per-platform tracking, defined prompt sets, share of voice by query cluster | One blended score, or screenshots |
| What’s your first 30 days? | Baseline audit, entity and mention gap analysis, prompt set agreed with you | Immediate content production |
| How do you earn third-party mentions? | Editorial PR, review platforms, community, named outlets | Vague “authority building” |
| What’s out of scope? | A clear list | Nothing — they do everything |
| What if the model describes us wrongly? | Correction workflow across sources the model uses | Not considered |
That last question matters more than it sounds. LLM answers frequently misstate pricing, positioning or product scope, and the fix usually lives in sources you don’t own. An agency that hasn’t thought about factual correction is optimising for presence without accuracy.
It’s also worth confirming your site is technically eligible. Google documents how its AI features access and cite content, and separate crawler permissions govern OpenAI, Anthropic and Perplexity. Blocking those agents in robots.txt while paying for citation work is a more common contradiction than it should be.
Who does what in this market?
- Panem. Runs AI visibility as a single programme — technical eligibility, entity clean-up, content restructuring and earned mentions — reported per assistant rather than as one blended score. Case study: twelve months with NewoldStamp, a B2B SaaS platform, took its AI audience to 7.7M a month, +49.8%, at ~1.3K citations monthly, with average organic position moving from 21.1 to 11.9. Best for ongoing programme work; overkill for a one-off audit.
- iPullRank. Technical bench with a documented point of view on how retrieval and relevance operate, applied to large or structurally complex sites. Case study: an AI search programme for a financial services brand reported a 120% lift in signups and 52% growth in organic traffic. Best for enterprise sites with real engineering problems; priced accordingly, and not a volume content shop.
- Siege Media. Content and digital PR specialists, which maps onto AI visibility given how many citations originate off-domain. Case study: a bottom-funnel strategy for Mentimeter generated roughly 250,000 ChatGPT visits. Best for brands whose gap is off-domain authority and LLM referral traffic; not the choice for heavy technical remediation.
- Omniscient Digital. B2B SaaS content strategy, with published research into where branded citations actually come from across the major assistants. Case study: Convert grew LLM visibility 81% and AI citations 140% inside 60 days. Best for software companies tying content to pipeline; narrow ideal client profile by design.
- Victorious. Runs a defined answer-engine optimisation line with fixed deliverables and structured reporting rather than an open-ended retainer. Case study: restructuring five pages for apparel brand UNIONBAY earned 84 AI Overview citations in six weeks. Best for mid-market teams wanting predictable scope; methodical rather than experimental.
Run your own baseline first
Run a baseline yourself first. Write twenty prompts a real buyer would type, run them across ChatGPT, Gemini, Perplexity and Claude, and record who gets cited and how you’re described. It takes an afternoon and it converts vendor conversations from theoretical to specific — you’ll know immediately whether a pitch addresses your actual gaps.
Then start narrow. A 90-day engagement on one product line, with an agreed prompt set and per-platform baselines, will tell you more than a twelve-month contract signed on a deck. For a fuller breakdown of how citation work is scoped and measured, Panem Agency publishes its methodology in more detail.
