AI marketing agents for outreach: How to earn mentions on the pages AI search cites

AI search has become one of the main places buyers go for information, at every stage of a purchase. They use it to make sense of a problem, to build a shortlist, and to choose between the last two vendors standing.

When an engine answers one of those questions, it doesn't rely on your website alone. In Profound's analysis of 11.84 billion AI citations, 43% pointed to pages brands don't control, like publications, review sites, forums, and video. That gives your outreach two audiences. The first is the editor you're trying to convince to mention you, add you to a comparison, or fix an outdated claim. Then there's every answer engine that reads the same page and folds it into what it believes about your brand. That belief decides whether you show up in answers, whether the facts about you are right, and whether the engines describe you favorably.

You still have to find the right pages and write a pitch worth answering. That part of the job hasn't changed. But you now also need to know which pages the engines cite, and for which prompts. No media database tracks that, and checking by hand, prompt by prompt and engine by engine, falls apart after a handful of topics.

Profound collects that data for you. It tracks which pages each engine cites, for which questions, and how that changes week to week. Profound also has Agents that take on the manual work of turning the data into target lists, pitches, and corrections. You can build your own, or start from Profound's library of agent templates.

Below, we walk you through the templates for each outreach job.

Find out how much outreach can influence your priority prompts

Before you build a list, find out whether outreach is the right lever for your topic at all.

According to our citation data, SaaS brands get about 11% of their citations from earned media, because the engines mostly quote vendors' own sites when they answer SaaS questions. Pharma and biotech brands get 59% of theirs from earned sources. Even within one industry, a company at the 25th percentile and one at the 75th can be 43 points apart in earned share.

So the same PR push can change a lot in one category and barely register in the next. If the engines answer questions on your topic by quoting vendor pages, a round of pitching won't do much. When they quote trade press instead, it could do a lot.

The Citation Type Breakdown for a Given Topic agent tells you which situation you're in. It pulls the pages the engines cite for a topic, sorts them by the kind of content they are, and gives you citation counts and share of voice for each group. You see at a glance how much of the answer comes from publications and review sites, and how much from vendors.

Build the list from the pages engines cite

Most media lists are ranked by reach. The outlets with the biggest audiences and the highest domain authority go at the top, along with the editors you already have a relationship with. For the people who'll read the coverage, that ranking still makes sense.

Answer engines don't work from your media list. When a buyer asks ChatGPT which project management tool suits a 50-person agency, the engine retrieves a handful of pages that answer that exact question and assembles its answer from them. Often that's a comparison post on a mid-size trade blog, a review site's category page, or a Reddit thread, and the same few pages tend to come up across many related questions. If one of those pages leaves you out, so does every answer built on it. Meanwhile, a feature in a much bigger publication that the engines never retrieve does little for you in AI search.

That's why the list you want is a list of specific pages, plus the people who can change them. The author of a cited comparison post can add you to it, and the editor of a cited review page can update your pricing. Three pre-built Profound agents help you build that list:

• The Create Media Target List agent pulls the earned media domains and pages cited in your category, with share of voice and citation counts for each, and analyzes which ones are worth pursuing.
• The Citation to Outreach Target Finder gets you from the page to the person. It takes the most-cited URLs for a topic, scrapes each page, finds the author's name, role, and contact details, and puts everything in a spreadsheet with one row per target.
• The Citation-Friendly PR Target Finder checks something most PR teams have never had to think about. It reads each cited domain's robots.txt and pushes outlets that block AI crawlers to the bottom of the list. If engines can't read a page, a mention on it won't make it into their answers, no matter how big the outlet is.

Write pitches that start from the page

A pitch written from the target page can ask for something specific. Maybe the comparison lists five vendors and leaves you out. Or the statistic in the second section is two years old, or the article describes a problem your product solves and never names a tool for it. Each of those changes makes the page better for readers, and that's why an editor says yes.

The Draft Outreach Email agent writes that kind of pitch. It scrapes the target page, reads it against your brand and product, and drafts an email around what the page says and what it's missing.

You could ask ChatGPT to write a pitch, too. It hasn't read the page, though, and it only knows your product as well as its training data or whatever documentation you feed it. If you launched a feature last month or rebranded last year, it may not know about either.

Correct what the web gets wrong about you

Answer engines describe you using the pages they retrieve. If some of those pages get it wrong, the engines are bound to repeat the mistake.

The mistakes come in two forms. Some are plain factual errors, like old pricing, a plan you retired, or an integration the page says you don't have. You can track those down and get them fixed. Others are about framing. A review that calls your onboarding painful, or a comparison that presents a competitor as the safer bet. Every fact on the page can be accurate, and the page can still pull the engines' view of you away from how you position yourself.

Profound's FactCheck handles the factual kind. It pulls the individual factual claims out of AI answers about your brand, checks each one against the product facts your team has given Profound, and traces every wrong claim back to the URL it came from. That last step is what makes it useful for outreach, because "ChatGPT has our pricing wrong" turns into a list of pages to contact. As for drafting, you have two agents that can help out:

• The FactCheck Action - Draft Outreach agent writes a correction request to the publisher of a page that repeats a wrong claim.
• The Sentiment Action - Draft Outreach agent starts from a third-party page that scored negative for sentiment. It isolates the negative claims, scrapes the live page, and drafts a short, factual email to the publisher.

If you'd rather hand corrections off entirely, FactCheck connects to Noble's Mention Refresh, which contacts the publisher and drafts the correction for you. FactCheck keeps checking the answers afterward, so you can watch your accuracy score climb as the engines pick up the fixed page.

WHOOP runs FactCheck across nine answer engines, and it flagged 7.9% of AI responses about WHOOP products for accuracy problems, from outdated specs to mixed-up product generations. "If an AI is telling someone something wrong about a WHOOP product, I can catch it, work out where it's coming from, and get a correction moving," says Kieran Riley, AEO Product Marketing Specialist at WHOOP.

Before you send anything about framing, check whether the complaint is about you or about your whole category. If the engines call every vendor in your space expensive, emailing one reviewer won't change that. A page of your own that makes the case for your pricing is the better bet.

Show up in the threads answer engines read

Forum threads are often cited in AI answers, and Google's AI surfaces, Gemini and AI Mode, cite social sources more readily than most other engines. A Reddit thread from years ago can still be feeding answers today, so the conversations happening in your category are worth joining. Profound gives you two agent templates that are Reddit-specific:

• The Reddit Comment Generator agent finds the most-cited Reddit thread on a topic in a subreddit over a recent period, and drafts a comment that fits the thread's tone and content.
• The FactCheck Action - Reddit Response drafts replies to threads carrying wrong or negative claims about your product.

The agents only draft. Someone from your company needs to post the comment from an account that makes the affiliation obvious and follows the subreddit's rules on self-promotion. Don't pose as a customer; communities are good at spotting it, and the thread where they call you out can easily become one more page the engines can cite about you.

See what changed with each new outreach run

In AI search, you can check two things after a placement goes live:

• Did the engines start citing the page for your topic?
• And do they describe you differently now?

Answer Engine Insights tracks citation share and sentiment by topic and by engine, so you can compare the weeks before a placement with the weeks after. The Top Sentiment Themes Report agent averages sentiment by theme for a topic over a period and tells you which themes got stronger and which got weaker.

You can also tag every outlet you pitch as one group in Profound and follow that group's citation share month to month. That shows whether the program as a whole is changing the answers, rather than one placement at a time.

Run outreach with your AI Marketer in Profound

Every agent above waits for someone on your team to run it. Before that, someone has to notice that a publication has started showing up in answers, or that a review page is still quoting last year's pricing, and decide it's worth a pitch. On a busy team, that noticing waits for a spare afternoon, and spare afternoons are rare.

AI Marketer (Aim) notices. It's the orchestrator inside Profound, a marketing teammate that reads your data every day and tells you the pitches you should make and why.

Aim works from two sources. The first is your AI search data, meaning which pages each engine cites, how the engines describe you, and which of their claims are wrong. Then there's Context Manager, where Profound stores your brand information, such as your positioning, audiences, competitors, products, approved claims, and current priorities. Your team fills it by connecting the places that knowledge already lives, like brand kits, knowledge bases, Slack, and Gong. Here's what a week looks like:

• Aim sends your team a ranked list of opportunities, each with the evidence behind it. For outreach, that might be a publication ChatGPT cites on your topic that has never mentioned you, a review page still quoting your old pricing, or a Reddit thread where the conversation about your product has turned sour.
• You pick which ones to act on. When you approve one, Aim turns it into a project, which is a plan broken into the specific tasks the opportunity needs. For the publication that has never covered you, the tasks might be pulling the pages it has published on your topic, finding each author's contact details, and drafting a pitch for each. Aim matches every task to an agent and runs them.
• Nothing goes out until your team approves it. All you have to do is review the output and ship it. Aim also learns from what you approve, edit, or turn down, so next week's list fits your priorities better.

You keep the relationships and the judgment about what's worth saying. Aim finds the pages influencing AI answers about you, tells you which to go after first, and has each pitch ready for your review.

Take Profound’s pre-built templates for a test drive or see what Aim can do for your outreach program. Book a demo.

AI marketing agents for outreach FAQs

Won't AI agents just automate spammy pitches at scale?

They would if you pointed them at a mass email list, so don't. Profound’s pre-built agents start from the handful of pages the engines cite for a topic and write each pitch from what that specific page says. The list is only as long as the set of cited pages, and every email is about its target.

How is a citation-based media list different from a media database?

Your media database tells you who covers your beat and how many people they reach. A citation-based list tells you which specific pages the engines pull from when buyers ask about your category, and who wrote them. The two overlap, but the citation list tends to include smaller publications, review sites, and individual articles that a database would rank near the bottom.

Should we still pitch high-reach outlets?

Yes, for all the usual reasons: readers, credibility, and something your sales team can share. Before you do, check whether the outlet shows up in AI answers on your topics and whether its robots.txt lets AI crawlers in. If it doesn't, the placement still reaches people. It just won't reach the answers.

How do I know if outreach changed what AI says about us?

Track citation share and sentiment on the topics you targeted, before and after the placement, engine by engine. Tag the outlets you pitch as a group and watch that group's citation share over time. That tells you whether the answers changed.

Does outreach matter more on some answer engines than others?

Yes. According to our citation data, ChatGPT draws 30% of its citations from earned media, more than any other major engine, while Gemini pulls 69% from brand-owned sites. If your buyers research in ChatGPT, give outreach a bigger share of the plan. Where Gemini is the bigger channel for you, start with your own pages.