How do AI assistants decide which businesses to recommend?

AI assistants recommend the businesses they can confidently cite, verify and act with. The strongest signals today are: being the source they already prefer, being easy to cite, plugging into their tools, and offering content with clear provenance.

If you want to influence those picks, prioritise three things: make your information easy to retrieve and quote, earn “preferred source” status where available, and integrate your first‑party data into the assistants’ tool ecosystems.

What signals actually move the needle in AI recommendations?

1) Citations: can the assistant back you up?

Citations are the public proof behind an AI answer. Modern assistants run their own retrieval, then cite the pages that best ground the claim—often via multi‑step, “fan‑out” sub‑queries that don’t mirror classic SEO rankings. Google explains this explicitly for AI Overviews and AI Mode, which may issue multiple related searches to build a response (Google Search Central).

Independent studies also show assistants pull from a wider, often different set of sources than traditional search. One analysis of AI Overviews found that nearly 30% of cited domains did not appear on the co‑displayed first‑page results, suggesting a distinct source‑selection process from Google’s standard ranking (arXiv study).

For ChatGPT’s web answers, OpenAI’s help docs confirm that results include citations you can click, and that its search module can issue additional targeted queries before finalising the answer (OpenAI Help).

What this means for you: you need pages that are crawlable, unambiguous and quote‑ready—clear claims, visible dates/authors, and plain‑language summaries that can be lifted into an answer card. (We break down the mechanics in: “Why your digital footprint determines whether AI recommends your business” on the Luciqo blog.)

2) Preferred sources: do users (and platforms) already “choose” you?

Google now lets users nominate preferred publishers; content from those sites can be highlighted with a badge inside AI Mode and AI Overviews. Google says people are twice as likely to click a Preferred Source and that more than 345,000 unique sources have been selected to date (Google product blog). Google also offers implementation guidance so publishers can attract “add as preferred source” selections (Search Central docs and a dedicated Preferred Sources guide).

Implication: if your audience opts you in as a preferred source, your odds of appearing (and being clicked) rise. Treat “preferred” as a new subscriber action alongside email sign‑ups.

3) Plugins, tools and first‑party data: can the assistant act with you?

Assistants increasingly privilege answers they can verify against connected tools or execute via APIs. Anthropic’s Claude documents web search and enterprise search that tie into connected sources such as SharePoint, Slack and Google Drive, returning well‑cited answers from those systems (Anthropic docs and Claude Help).

Perplexity’s Pro Search lets users pick models and always returns direct source links; it’s positioned for research‑grade transparency (Perplexity Help).

As these ecosystems mature, businesses that expose accurate pricing, specs, inventory, documentation and policies via well‑structured, tool‑readable pages and APIs gain an edge because assistants can check, cite and act.

4) Provenance and safety: can the assistant trust the asset?

Content provenance standards like C2PA (“Content Credentials”) and watermarking such as SynthID are being adopted across the industry. Google says it uses Content Credentials across generative media tools and has added SynthID verification in the Gemini app (Google blog). OpenAI documents how its APIs report C2PA/SynthID signals for images, and notes that some transformations can strip these signals (OpenAI dev docs). The C2PA technical spec itself details how tamper‑evident manifests establish trust in assets (C2PA 1.4 spec).

While provenance isn’t (yet) a universal ranking factor, assets with clear authorship, dates and origin signals are safer to cite—especially on sensitive topics.

How do major assistants choose recommendations today?

ChatGPT (OpenAI)

ChatGPT’s web answers show citations and may refine results with additional queries before it decides what to cite (OpenAI Help). Independent observers have also found that source selection can vary with hidden retrieval “pipelines,” making results—and citations—shift as OpenAI rotates providers, which underlines the need for broadly readable, well‑structured content (Search Engine Land).

In practice: pages with clean HTML, explicit facts and clear authorship/dates tend to get cited more reliably across assistants—because the retrieval layer can extract them with high confidence.

Google: AI Overviews and AI Mode

Google’s AI features cite supporting sources and may run multiple sub‑queries to assemble answers (AI Overviews explainer; Search Central docs). Google is also rolling out UX updates to make the linked sources more prominent in AI Mode and AI Overviews (Android Central). And users can set Preferred Sources, boosting visibility for selected publishers (Google product blog).

Takeaway: if you’re a brand or agency, your job is to earn inclusion in those support links and to encourage your audience to add you as a Preferred Source.

Perplexity

Perplexity positions itself around transparency: every answer includes direct links, and Pro Search lets users choose the model for the session (Perplexity Help). Studies of citation behaviour across engines consistently show Perplexity at the “always cites” end of the spectrum; as a result, it can be a strong early indicator of which of your pages are most quote‑able.

Practical angle: seed pages with concise definitions, comparison tables and spec sheets that directly answer high‑intent questions. Those blocks are what Perplexity is most likely to lift and link.

Claude (Anthropic)

Claude’s web search and enterprise search features fetch current content and return answers with citations; in enterprise contexts it can search across connected internal tools and repositories (Anthropic platform docs; Claude Help). On the public web, it behaves similarly to others: clear, well‑structured sources with verifiable claims and dates earn more citations.

Which parts of your digital footprint should you prioritise?

• Source pages designed for citation: include a one‑paragraph summary, bullet points of key facts, a dated author byline, and a plain‑English H1 that mirrors the user question. This aligns with how assistants assemble snippets and justify links (OpenAI Help; Search Central AI features).
• Preferred Source readiness: add an on‑site prompt to “Add us as a Preferred Source on Google,” and follow Google’s guidance so your root domain is eligible (Google product blog; Preferred Sources guide).
• First‑party data exposure: publish pricing, features, specs, availability and policies on crawlable URLs; where possible, offer an API or machine‑readable feeds so assistants and tools can verify and act (Anthropic web/enterprise search).
• Provenance cues: add visible dates/authors and, for media, embed C2PA Content Credentials where your stack supports it. It helps downstream systems and “About this image”‑style features display trustworthy origin info (C2PA spec; Google Images metadata note).

If you want to see how assistants currently describe—and rank—your brand and competitors across buyer prompts, we built Luciqo to capture exactly that and link it to revenue for UK teams. You can learn more here: luciqo.ai.

A 90‑day plan to get noticed by assistants

I’ve used this plan with agency and in‑house teams. It’s focused, measurable and realistic for a quarter.

Days 1–14: baseline and fix what’s broken

• Run an AI visibility baseline: for your top 25 buyer questions, capture how ChatGPT, Gemini, Claude and Perplexity answer, who they recommend, and which sources they cite. Tag by intent (comparison, pricing, alternatives, best‑for…). If you prefer not to build a tracker, Luciqo automates this and joins the dots to GA/HubSpot.
• Identify “closest‑to‑quote” pages: which of your pages are already being cited, or nearly cited (e.g., competitors cited for a topic you cover)? Prioritise 10.
• On those 10 pages, ship fast edits: add a one‑paragraph plain‑English summary, a date/author line, a short FAQ, and a comparison block. Ensure prices/specs are explicit and current.
• Encourage Preferred Source adds: add a call‑to‑action on your site and in email to “Add us as your preferred source on Google,” following Google’s guidance (Google product blog).

Days 15–45: create 6 citation‑magnet assets

• Publish six short, high‑intent pages that map exactly to buyer prompts (e.g., “Best [category] for [persona] in the UK”, “[Brand] vs [Competitor]”, “Pricing: [Product]”). Each page should include: 100–200‑word summary, bullets of facts, a small spec table, and a dated author line.
• Add structured data that reflects what’s visible on the page; don’t invent special files for Google’s AI features—Google says you don’t need LLMs.txt or similar for inclusion (Search Central AI optimisation guide).
• Expose first‑party facts wherever possible: coverage zones, SLAs, security certs, integration lists, stock/lead‑time windows. These are the details assistants lean on to justify recommendations.
• For media‑heavy pages, embed C2PA Content Credentials (supported across Adobe tools) so provenance survives downstream (Adobe Experience League).

Days 46–75: tool connections and proofs

• Connect internal sources that assistants can search (help docs, implementation guides, status pages). If you’re piloting Claude in the org, set up enterprise search across your repositories so agents can cite your own documents back (Claude Help).
• Publish a public “How we compare” asset (clean HTML, no heavy scripts) and keep it updated monthly. Assistants often lift these blocks when answering “best” or “alternatives” prompts.
• Launch a lightweight pricing and spec API or a downloadable CSV that mirrors your web facts. Assistants that can verify via tools tend to prefer verifiable answers.

Days 76–90: measure, tighten, repeat

• Re‑run your 25‑prompt test and compare citations and recommendations vs. Day 1. Track shifts in engines and in Google AI features (Overviews/Mode). Google’s own docs describe how AI features contribute to Search Console and how to analyse traffic changes alongside Analytics (Search Central).
• Double down on what’s working: create two more assets modelled on your highest‑cited page. De‑prioritise formats that aren’t earning citations.
• Put a calendar reminder to refresh dates/authors and fact blocks. Stale pages lose trust and are less likely to be cited.

How do I influence each signal, practically?

• Citations: add a “TL;DR” paragraph and fact bullets near the top; use descriptive sub‑heads that mirror buyer questions; include a visible publish/update date; avoid burying numbers in images.
• Preferred sources: promote the “Add as preferred source” action across your newsletter and footer; ensure the root domain (not just a subdirectory) represents your best work (Preferred Sources guide).
• Plugins/tools: maintain a public page for status, pricing, integrations and SLAs; ship a simple JSON/CSV endpoint; document usage clearly so assistants can quote and verify.
• Provenance: embed Content Credentials for media where possible; add author bios and link to policy pages for sensitive claims (C2PA spec).

For deeper context on why rankings alone no longer guarantee traffic, and how assistants reshape the buyer journey, see these Luciqo posts:

• Why High Google Rankings No Longer Guarantee Website Traffic
• Your Brand Ranks on Google. So Why Isn’t AI Recommending It?
• Why your digital footprint determines whether AI recommends your business

Quick checklist for marketing leaders

• Do we have 10+ quote‑ready pages with clear facts, dates and authors?
• Can we encourage customers to add us as a Preferred Source on Google?
• Are our prices, specs and policies published in clean HTML and a machine‑readable format?
• Do our key media assets carry provenance (C2PA) where feasible?
• Are we measuring our brand’s mentions, citations and recommendations across ChatGPT, Gemini, Perplexity and Claude?

What’s changing next—and what to watch

Two shifts matter for Q4 2026 planning. First, assistants are getting better at showing and weighting sources. Google says it’s rolling out updates to highlight linked sources more clearly in AI Mode and AI Overviews (Android Central). Second, provenance is spreading fast: Google, OpenAI and Adobe are all pushing Content Credentials and watermarking through their products (Google blog; OpenAI dev docs; Adobe Experience League).

Closing: start with the pages assistants can safely cite

If you remember one thing from this piece, make it this: assistants don’t merely “rank” sites—they assemble answers they can defend. That means clear, current, provenance‑backed facts on pages designed to be quoted. Start there, build Preferred Source momentum, connect your first‑party data, and measure monthly. If you want a single view of how assistants describe and recommend your brand today, we built Luciqo to help you test, track and tie those gains to revenue.

Frequently asked questions

Which assistants show the most citations today?

Perplexity consistently links to sources in every answer and lets users choose the model (Pro Search), while ChatGPT, Gemini and Claude also show citations when web search is invoked. See Perplexity’s help docs and OpenAI/Anthropic documentation for details.

Does Google’s Preferred Sources really help brands get recommended?

Google reports users are twice as likely to click a Preferred Source, and more than 345,000 unique sources have already been selected. It’s worth nudging your audience to add you.

Do I need special files like LLMs.txt for Google’s AI Overviews?

No. Google’s own guidance says you don’t need new machine‑readable files like LLMs.txt to appear in AI features. Focus on accurate, crawlable content and matching structured data to visible text.

What provenance steps are practical right now?

Add visible dates/authors on all key pages. For media, embed C2PA Content Credentials (supported across Adobe tools). Know that some transformations can strip watermarks/metadata, so keep originals archived.

How do I measure whether assistants recommend my brand more often?

Track two things monthly: the share of answers that mention or recommend you for your top 25 buyer prompts, and which of your URLs are cited. Luciqo automates this and connects it to GA, Matomo and HubSpot.

What should I prioritise in the first 30 days?

Fix your top 10 “almost there” pages: add a TL;DR, bullets of facts, up‑to‑date pricing/specs, a comparison block, and a dated author line. Then ask your audience to add you as a Preferred Source on Google.