A pitch list ranked by what the models already quote
Most of what AI answers say about a brand does not come from that brand. It comes from review platforms, community threads, comparison posts and press: sources the model already trusts and already reads. When your own pages are in order and you are still absent, the work moves off your domain, and the useful version of it starts from citation data rather than a generic outreach list.
That changes what the outreach team is optimising for. Link building chased a ranking signal; this chases whether a model quotes a source when it answers a buyer's question, and how that source characterises you when it does. The pitch list is different, the success measure is different, and the timeline runs on the engines' re-crawl schedule rather than the publication date.
Why comms and off-page teams work this way
Target the sources with evidence, not intuition
Citation data shows which domains each engine quotes for the prompts your buyers ask. That is a pitch list ordered by demonstrated influence on the answer rather than by domain authority. A trade publication with modest reach can be cited on every category question while a national title is never quoted.
Coverage becomes measurable
A placement either changes how often the models cite you on the prompts it was meant to move, or it does not. That is a harder outcome measure than impressions or a clipping report, and one off-page teams have rarely been able to produce.
Community sources count
Reddit, Quora and niche forums are quoted heavily by several engines, particularly on "is X any good" questions. They need a slower, participation-led approach than a press pitch, they punish anything that reads as marketing, and they need ongoing monitoring rather than a one-off push. Treating them as a placement target is the fastest way to be removed from them.
Accuracy is a bigger lever than volume
A cited source that describes you wrongly, with stale pricing, a discontinued plan or a competitor's feature list, does more damage than ten sources that omit you. Because citation data names the source behind an inaccurate answer, correcting it is a specific request to a specific publisher rather than a general reputation campaign.
How off-site authority building works
Turning citation data into an outreach programme whose results you can actually attribute.
- 1
Pull the cited-source list for your category
Run the prompts your buyers ask across the engines that matter and collect every domain and URL the engines attach as a source. Do it per engine, because Perplexity, browsing ChatGPT and Google AI Overviews lean on noticeably different source mixes. Segment the list into editorial, review platform, community and owned, and count how many prompts each source appears on.
- 2
Score the gap on each source
For each source, record whether you appear at all, whether a named competitor appears, and how the source describes you if it mentions you. Rank the list with inaccurate mentions first, then absences on high-frequency sources, then the rest. A cited source that describes you inaccurately is more urgent than one that omits you, because it shapes the answer in the wrong direction every time it is fetched.
- 3
Match the source type to a workflow
Editorial needs a story and a pitch. Review platforms need a complete profile, current screenshots and a steady flow of genuine reviews prompted at the right moment. Communities need weeks of real participation from people who know the product, answering where the brand is relevant and staying quiet where it is not. Owned syndication needs a canonical strategy so the copy does not compete with your own site for the citation.
- 4
Place, then wait for re-crawl
A new mention only affects answers once the engine re-fetches the source. Retrieval-first engines pick it up fastest, browsing ChatGPT follows, and anything answered from training data will not move until the model is refreshed. Note the date each placement went live, watch crawler activity on the source where you can, and set the expectation with stakeholders before the placement rather than after.
- 5
Re-measure against the same prompts, and keep the list live
Compare mention rate, citation rate and sentiment on the baselined prompt set, per engine, once a re-crawl has had time to happen. Attribute the change to a placement only where the engine has started citing that specific source, which citation data shows directly. Then re-pull the cited-source list quarterly, or monthly in an active category, and treat new entrants as new targets.
How teams use it
Digital PR with an outcome metric
Pitch the publications that demonstrably feed AI answers in your category, and report the citation change rather than the clip count. The list is smaller and more specific than a general media list, and the result is a movement in the answers rather than a collection of links.
Review-platform strategy
G2, Capterra, Trustpilot and their category equivalents are quoted constantly in recommendation answers. Profile completeness, review recency and accurate category placement are an AI-visibility lever, not only a sales asset, and usually the cheapest source on the list to improve.
Community presence
Find the threads and subreddits that already surface in answers about your category and build a genuine presence there over weeks. The monitoring side, seeing which threads are cited and how your brand is discussed in them, is covered by Reddit Intelligence.
Correcting an inaccurate narrative
When models repeat something wrong, the fix is changing the sources they read, not the page you wish they read. Citation data names the source, which turns the job into a correction request to a specific publisher or a profile update on a specific platform.
Launch coverage that reaches the models
A launch generates press, but press the engines never cite does not change what buyers are told. Checking which launch coverage was picked up as a source, and which was not, tells you whether the announcement reached the answer or only the news cycle.
What differs per engine
The source mix differs by engine, so the pitch list does too.
ChatGPT
When ChatGPT browses, it fetches via OAI-SearchBot and ChatGPT-User and shows sources inline, so a new placement is visible as soon as the source is re-fetched. Its mix leans on established editorial and review domains. When it answers from training, coverage published after the cutoff has no effect until the next model update.
Perplexity
Perplexity is the most community-heavy engine and the fastest to reflect a new source. Reddit threads, niche forums and recent comparison posts appear in its citations often, and PerplexityBot re-fetches frequently. It is the best engine for testing whether a placement has any effect, and where community participation shows up first.
Google AI Overviews / AI Mode
AI Overviews draw on Google's index, so the cited sources are typically pages that already rank for the query: editorial roundups, review platforms and indexed forum threads. A placement helps here only once Google has indexed it and only if the page is competitive for the query. Google-Extended is irrelevant to this surface.
Gemini
Gemini grounds on Google Search and, for product questions, on merchant data, so the same indexed sources apply with review and shopping platforms weighted more heavily. Source display varies by surface, so tracking through the platform rather than by hand is the only reliable way to see which placements it has picked up.
The numbers to report
How to report off-site work in figures the engines will bear out.
- Cited-source coverage
- Of the sources cited for your prompt set, the share that mention you at all. This is the primary gap measure. Track it per source type so the report shows whether the gap is editorial, review, community or syndication.
- Mention accuracy
- The share of cited sources that describe you correctly. An inaccurate mention on a frequently cited source is the most urgent item on the list, and fixing it usually moves the answer faster than a new placement because the engine is already reading that page.
- Citation rate by source
- For each placement, whether the engines have started citing that specific URL, per engine. This is the direct evidence a placement reached the answer. A placement that is live but never cited was on the wrong source or was never fetched, and the crawler log tells you which.
- Sentiment shift
- How the engines characterise you before and after, on the same prompts. A placement that raises mention rate but introduces a negative characterisation is not a win. Track tone per engine, since engines reading different sources describe you differently.
Common mistakes and the fix
Mistake · Pitching from a domain-authority list
Fix · Pull the cited-source list from the engines and rank it by how many prompts each source appears on. Domain authority predicts rankings; it does not predict what a model cites, and the two lists overlap less than teams expect.
Mistake · Judging a placement in its first week
Fix · Log the placement date, note which engines have re-fetched the source, and re-measure monthly. Retrieval engines move in days; training-grounded answers may take months. Set that expectation before the placement goes live.
Mistake · Treating Reddit as a placement channel
Fix · Participate, do not pitch. Have people who know the product answer real questions over weeks, stay out of threads where you are not relevant, and monitor how the brand is discussed. Astroturfing is detectable and gets accounts removed.
Mistake · Ignoring the sources that already mention you
Fix · Score existing mentions for accuracy before chasing new ones. An outdated description on a source the engines already cite does more harm than any absence, and a correction request is faster than a new pitch.
A worked example: a B2B payroll software vendor
A hypothetical payroll software vendor has a server-rendered site, complete schema and open crawler access, and is still absent from ChatGPT and Perplexity answers to "best payroll software for small businesses". The cited-source list for a fifteen-prompt set shows two review platforms, three comparison blogs and a small-business forum accounting for most citations, with the vendor's own site cited only on brand-name prompts.
Scoring the gap shows the review-platform profile is incomplete, has no reviews from the past year and sits in a broader category than the one buyers ask about. One comparison blog lists pricing two years out of date. The forum has threads recommending competitors where the vendor is never mentioned. The list is ordered accordingly: correct the pricing first, complete the profile and prompt recent customers for reviews second, begin genuine forum participation third, and defer editorial pitching.
The pricing correction is one email to the blog's editor, and Perplexity re-fetches the updated post within the month, after which the stale figure disappears from its answers. The profile category change is immediate; reviews arrive over two months as customers are prompted at renewal. At the quarterly re-measure, Perplexity cites the updated profile and names the vendor on most category prompts, browsing ChatGPT on some, and the forum threads are not yet cited, which the team reads as expected rather than as failure.
Frequently asked questions
Is this link building?+
No. Link building optimises for a ranking signal, and the link is the unit of value. This optimises for whether a model quotes a source when answering a question, and how that source characterises you. An unlinked mention in a cited source can move an answer.
Should we pay for placements?+
No. Paid links and sponsored placements carry search penalties and, more practically, tend to sit on sources the models do not cite. Earned coverage on genuinely cited domains is the only version of this work that compounds.
How do community sources like Reddit fit in?+
They are cited heavily, and they punish anything that reads as marketing. Treat them as a participation channel with a long horizon, have people who know the product answer real questions, monitor how your brand is discussed, and never astroturf. It is detectable, and a removed account takes its genuine history with it.
How do we prove a placement worked?+
Baseline mention rate, citation rate and sentiment on a fixed prompt set before the placement, then re-measure after the source has been re-crawled. Citation data shows whether the engine has started citing that specific source, which is the direct link between placement and answer.
Which source types matter most?+
It depends on the category and the engine, which is why the list has to be pulled rather than assumed. Software categories lean on review platforms and comparison posts; consumer categories on editorial roundups and community threads; technical questions on documentation. The list for your prompts will show which applies.
How long does a placement take to show up?+
From days to months depending on the engine. Perplexity re-fetches frequently and reflects new sources quickly; browsing ChatGPT follows; answers grounded in training data will not change until the model is refreshed. Log the placement date and read results per engine rather than expecting one timeline.
Can we syndicate our own content to build authority?+
Yes, with care. Republishing on partner sites can put your claims on domains the engines already trust, but without a canonical strategy the copy competes with your own page for the same citation, and the partner may win. Use canonical tags where supported and vary the copy where not.
Who should own this internally?+
Usually comms or PR, with SEO supplying the citation data and the technical checks. The workflows are pitching, profile management and community participation, which are comms skills. The prompt set should be agreed by both teams so the report means the same thing to everyone.
Explore the features behind this solution
Go deeper
See which sources AI quotes about your category
Start from the domains that already shape the answer, and pitch those.