The short answer.
AI engines that search the web recommend B2B companies largely from third-party pages: review platforms such as G2 and Clutch, “best X” comparison lists, industry directories, Reddit and other communities, Wikipedia, LinkedIn, YouTube and the press. Each engine leans on a different mix, so a company needs an accurate, genuine presence on several source types, not just a good website.
The work is not mysterious, and it cannot be bought honestly in bulk. It is mostly correcting what already exists about you, asking real clients for real reviews, and giving editors and communities something worth citing. This guide catalogues each source type and what the evidence says about it.
- 01 · REVIEWSG2, Clutch, Capterra, Trustpilot
- 02 · LISTS“Best X” roundups and comparisons
- 03 · DIRECTORIESIndustry and partner listings
- 04 · COMMUNITYReddit, forums, Q&A sites
- 05 · REFERENCEWikipedia and Wikidata
- 06 · MEDIAPress, podcasts, LinkedIn, YouTube
Six source types, covered one by one below.
Ten sources, none of them ours.
On 6 November 2026 we asked Perplexity, logged out, which agencies are best for improving a B2B company’s visibility in AI search in the United States. It named ten agencies, built from ten sources. We were not among the names, and none of the ten sources was ours.
The sources are the lesson. Two were pages from one SEO-software company’s agency directory. One was a press-release distribution site, one a finance news portal and one a business publication. Five were blogs published by agencies or marketing-technology companies. Perplexity itself warned in the answer that many such lists are rankings written by agencies. The engine knew the sources were weak, and used them anyway, because they were what existed.
Perplexity, logged out, default search, 6 Nov 2026. Classified by us from the source domains.
The other engines told the same story. ChatGPT named eight agencies and the Google AI Overview six; we were in neither list, and only one agency appeared in all three engines. We have written up why that happens in why your brand is missing from AI shortlists; read that for the diagnosis. This article is the catalogue of where to show up instead.
What the studies say engines read.
Independent, peer-reviewed research on AI citations is still thin. Most of what exists comes from companies that sell AI visibility tools, so treat every number as a direction, not a law. The largest we found is Profound’s citation study, updated August 2025, covering August 2024 to June 2025 across ChatGPT, Google AI Overviews and Perplexity.
| Engine | Most-cited domains | What it suggests |
|---|---|---|
| ChatGPT | Wikipedia 7.8%, Reddit 1.8%, Forbes 1.1%, G2 1.1%, TechRadar 0.9% | Reference and established publishers weigh heavily |
| Google AI Overviews | Reddit 2.2%, YouTube 1.9%, Quora 1.5%, LinkedIn 1.3%, Gartner 0.7% | Community, video and professional networks matter |
| Perplexity | Reddit 6.6%, YouTube 2.0%, Gartner 1.0%, Yelp 0.8%, LinkedIn 0.8% | Communities first, then analysts and review sites |
Two cautions. First, these are shares of all citations across every topic, so no single domain dominates; the long tail is enormous. Second, the mix shifts over time and differs by industry, so a software buyer’s question will draw on different sources from a question about restaurants.
A second, larger signal comes from Ahrefs. Its December 2025 study of 75,000 brands found that branded web mentions correlated with AI visibility at between 0.66 and 0.71 across ChatGPT, Google AI Mode and AI Overviews, while link metrics correlated only weakly. That is a correlation, not proof of cause, but it points the same way: being talked about on other sites matters more than the number of links to yours.
Review platforms: G2, Clutch, Capterra, Trustpilot.
For B2B software and services, review platforms are the clearest overlap between what buyers read and what engines cite. G2 appears among ChatGPT’s most-cited domains in the Profound data. And in G2’s own survey of more than 1,000 B2B software buyers, published 6 October 2025, half said they now start their buying research in an AI chatbot rather than Google, with AI chat ranked first among sources shaping their shortlist and review websites second. G2 sells to vendors, so read that with care, but it matches what we see.
Which platform for which business
Software companies are usually expected on G2 and Capterra. Agencies and professional services are more often compared on Clutch. Trustpilot is broader and more consumer-facing. Pick the one or two where your buyers already compare providers; a thin profile on five platforms helps less than a complete one on two.
What to do this week
- Claim every profile that exists in your name, and correct the description, founding year, location and services.
- Ask three recent clients, by name and by email, for an honest review. Do not offer anything for a positive one.
- Reply to every review, including the critical ones, briefly and factually.
Our own Clutch profile carries a verified review from Leopard Project Controls, 5.0 overall, published 3 July 2024, for a project that ran from January 2022 to March 2024. One genuine, detailed review on the right platform is worth more than ten thin ones.
The rules here are strict. The US Federal Trade Commission’s rule announced on 14 August 2024 bans fake reviews, including AI-generated ones, buying reviews you should know are fake, paying for reviews only if they are positive, and undisclosed reviews from your own staff.
“Best X” lists and roundups.
Comparison lists are the source type engines use most for “which provider should I choose” questions, and the one most open to abuse. An Ahrefs study published 4 December 2025 analysed ChatGPT’s answers to 750 prompts about software, products and agencies and found that “best X” blog lists made up 43.8% of all page types cited. In 250 Google results for “best X software”-style searches, 169, or 67.6%, featured a list in which the company writing the article ranked itself first.
Peec AI, which sells AI search analytics, measured how much of this self-ranking content engines actually cite. Across about 232,000 citations from December 2025 to February 2026, roughly 11% came from listicles in which the publisher recommended its own product. The rate differed sharply by engine: about 3.6% for ChatGPT, against about 10.3% for Google AI Mode and 10.4% for Perplexity. Peec’s data covered software review prompts only.
The honest way into lists
- Find the lists engines already cite for your category by running your buyers’ questions and reading the sources.
- Write to the editors of independent lists with a short, factual note: what you do, for whom, and one dated result they can check.
- If you publish your own comparison, say clearly that it is yours, include genuine competitors fairly, and date it.
Do not pay for a place on a list presented as independent without that payment being disclosed. The FTC’s 2024 rule specifically names company-controlled review sites presented as unbiased, and an engine that learns to discount such lists will discount you with them.
Industry directories and partner pages.
Directories are cheap to get right and quietly influential. Two of the ten sources Perplexity used in our test were pages from a single agency directory run by an SEO-software company. Directories also explain one of the errors engines made about us: asked about Prolinkage by name, ChatGPT mixed up our founding year, because our own profiles say 2015 and some older directories say 2011. The engine was not wrong to read those directories. We were wrong to leave them stale.
Partner pages work the same way. If you are a certified partner of a platform your buyers use, a listing on that platform’s partner finder is a third-party page confirming what you do. It is also checkable, which is the point.
What to do this week
- Search your company name plus “directory”, “profile” and “reviews”, and list every listing you find.
- Correct the founding year, location, size, services and website link on each, so every source says the same thing.
- Ask each platform partner whether you qualify for their public partner listing.
Reddit and communities.
Reddit is the single most-cited domain for Google AI Overviews and Perplexity in Profound’s data, and second for ChatGPT. Measured within each engine’s ten most-cited sources, Reddit took 46.7% of Perplexity’s and 21% of Google AI Overviews’. For B2B buyers, the relevant places are specialist subreddits, professional Slack and Discord groups, Quora and industry forums where people ask peers which provider to use.
This is also where shortcuts do the most damage. Communities spot planted recommendations quickly, and most have rules against self-promotion. The only approach that lasts is participation: named people from your company answering questions in their field, saying who they work for, and recommending a competitor when that is the right answer.
What to do this week
- Search Reddit and Quora for your category plus “recommend” or “agency”, and read what buyers ask.
- Pick one community where your buyers are active, read its rules, and have one expert answer one question a week, openly.
Wikipedia and Wikidata.
Wikipedia is ChatGPT’s most-cited domain in Profound’s study, at 7.8% of all its citations and 47.9% of citations within its top ten sources. That makes it tempting. For most $1M to $50M companies it is also the wrong place to start. Wikipedia only covers organisations that independent, reliable sources have already written about in depth, and its conflict-of-interest guideline says people with a connection to a subject are “strongly discouraged from editing affected articles directly.” Paid editors must disclose who is paying them, and changes should be proposed on the article’s talk page.
So the order is reversed. Earn independent coverage first, through press and genuine expertise. If a Wikipedia article about your company or your category already exists and contains errors, propose corrections on its talk page with sources. Wikidata, the structured database behind Wikipedia, can hold basic facts such as your founding date and website. We found no rigorous study showing how much engines rely on it for B2B recommendations, so treat it as a low-effort accuracy check, not a lever.
Press, podcasts, LinkedIn and video.
Established publishers appear among ChatGPT’s most-cited domains in the Profound data, LinkedIn among Google AI Overviews’ and Perplexity’s, and YouTube is the second most-cited domain for both of those engines. For a B2B company these are the channels where a named expert can be quoted, interviewed or watched explaining their work.
| Source | What earns a place | What to avoid |
|---|---|---|
| Trade and business press | Original data, a clear opinion, a client story told with permission | Paid articles presented as editorial without a disclosure |
| Press-release sites | Factual announcements: launches, partnerships, research | Treating a release as coverage; nobody independent wrote it |
| Podcasts | A guest with a specific, useful story for that audience | Pay-to-appear shows with no real audience |
| A complete company page and experts posting dated evidence | Engagement groups and borrowed posts | |
| YouTube | Short explainers that answer one buyer question each | Recycled slide decks with no voice or evidence |
A press-release distribution site did appear among Perplexity’s sources in our test, so releases can be read. But a release is your own words on someone else’s domain. Use it for announcements, and do not count it as independent proof.
Effort against likely value.
Not every source type deserves your first month. This matrix is our judgement, informed by the studies above, not a measured result. Value means how likely the source is to be read by engines and buyers for a typical B2B category question; effort means time and access, not money.
- Correct existing profiles and directories
- Partner listings
- Genuine reviews on one or two platforms
- Inclusion on independent “best X” lists
- Expert answers in communities
- LinkedIn company page and expert posts
- Press and podcasts
- Wikipedia (only if already notable)
- Wikidata facts
- Press-release distribution
The top-left corner is where most companies should spend their first fortnight. It is unglamorous: logins, descriptions, founding years. It is also the work that stops engines repeating wrong facts about you.
The off-site footprint audit.
Tick each statement that is true for your company today. You get a coverage score for each source type and an overall score. It takes about fifteen minutes with a browser open. Nothing you tick is sent anywhere.
Footprint coverage: 0%
Tick what is true today. The lowest bar is where to start.
If your lowest bar is directories or reviews, that is good news: those are the cheapest gaps to close.
A 90-day plan that stays honest.
Off-site work goes wrong when it is treated as a volume target. The plan below is built to be defensible if a buyer, a journalist or a platform ever looks closely at how your presence was earned.
- WEEK 1–2 · BASELINERun 20 buyer questions; record names and sources
- WEEK 2–4 · TIDYCorrect every profile and directory
- WEEK 3–6 · REVIEWSAsk real clients on one or two platforms
- WEEK 5–10 · LISTSPitch independent lists with dated proof
- WEEK 6–12 · VOICECommunity answers, press, LinkedIn
- WEEK 12 · RE-TESTSame questions, same engines, compare
Each step uses the sources the baseline showed, not a generic list.
The lines we do not cross
- No reviews written by staff, friends or paid services, and no reward for a positive review.
- No paid placement on a list or in an article unless it is disclosed as paid.
- No anonymous accounts recommending ourselves in communities.
- No direct editing of Wikipedia articles about ourselves.
- No claim about a client without their permission and a dated source.
Because AI answers vary from run to run, judge the re-test on how often you are named across several runs of each question, not on one screenshot. Our guide on how to check your AI search visibility shows how to set up that test, and the AEO and GEO visibility guide covers the on-site half of the work. When you start being named, our article on turning AI visibility into qualified enquiries covers what to measure next.
Questions owners ask.
What sources does ChatGPT use to recommend B2B companies?
When it searches the web, ChatGPT draws on reference sites, publishers, review platforms and comparison lists. Profound’s 2024–2025 study found Wikipedia its most-cited domain, with Reddit, Forbes and G2 also prominent, and an Ahrefs study found “best X” lists were the most common page type in its sources.
Do directories help with AI search?
They can. Engines read directory pages when comparing providers, and in our own test two of Perplexity’s ten sources were directory pages. Just as important, stale directories feed engines wrong facts. Correct every listing so your founding year, location and services match everywhere.
Which review sites matter most for B2B AI visibility?
The ones your buyers already use to compare providers. Software companies are usually expected on G2 and Capterra; agencies and services on Clutch. A complete profile with genuine recent reviews on one or two platforms does more than thin profiles on many.
Can we pay to be listed in AI answers?
Not in the organic answer itself. You can pay for some directory placements or list entries, but undisclosed paid placement presented as independent risks breaching the US FTC’s 2024 rule and platform terms, and it is the kind of source engines may learn to discount.
How long does off-site work take to show up?
Profile corrections can be picked up as soon as engines next read those pages. Reviews, list inclusions and press take months to accumulate. Re-test the same buyer questions after about twelve weeks and compare how often you are named across several runs.
Sources: Prolinkage’s own AI engine test, 6 November 2026 (Perplexity, ChatGPT, Google AI Overview). Clutch verified review, Leopard Project Controls, 3 Jul 2024. External: Profound, AI platform citation patterns, updated Aug 2025; Ahrefs, brand visibility correlations, 12 Dec 2025; Ahrefs, “best” lists research, 4 Dec 2025; Peec AI, self-promotional listicles, 3 Mar 2026; G2, AI chat and B2B software buying, 6 Oct 2025; US FTC, fake reviews rule, 14 Aug 2024; Wikipedia, conflict-of-interest guideline. All accessed 7 November 2026.