How do I get onto the pages AI already cites?

Write down the questions your buyers ask. See which pages ChatGPT and Perplexity cite for them. Mark the pages that name other businesses but not yours. Then act on each kind of page the way that page allows. It is slow work done by hand. A cited page shows where an answer looked. It does not show why the answer named someone. So doing all of this can still leave you unnamed.

This post is the “how” that follows our post on which websites AI cites. That post covers which kinds of sites show up. This one covers what to do about them, using Answersky’s own tests.

Why look at the pages AI already cites?

They show where the assistant looked before it answered. When an assistant searches the web it often points to the pages it used. Being on one of those pages is not the same as being named in the answer. That is why we say “cited” and not “recommended”.

In Answersky’s Sep 30 test answersky.com was cited in 0 of 54 answers. In Answersky’s Oct 7 test it was 0 of 36. That tells us our own site was not a source. It does not tell us whether a page on another site mentions us. If you want the reasons an assistant might skip a business, read why ChatGPT doesn’t recommend my business.

What do I need before I start?

Your buyers’ questions written the way a buyer types them before they know your name. Write 5 to 10. In Answersky’s Sep 30 test we used 9 questions. We put them to ChatGPT and Perplexity through their APIs with web search so the apps themselves may answer differently.

No list yet? How to check if ChatGPT mentions your business shows how to write the questions and keep a sheet.

How do I see which pages are cited?

Ask Perplexity and ChatGPT each question in a fresh chat. Copy the full address of every source they show and not only the site name.

Keep the page address because the pages behind a site change. In Answersky’s Sep 30 test LinkedIn was cited in 9 of 54 answers. In Answersky’s Oct 7 test it was 10 of 36. The pages behind those counts were not the same. In Answersky’s Sep 30 test the answers cited 12 different LinkedIn pages. In Answersky’s Oct 7 test they cited 5. In Answersky’s Oct 7 test only 2 of those 5 pages had also been cited on Sep 30. A list of sites is a start. A list of pages is something you can act on.

How we counted. In Answersky’s Sep 30 test we asked 9 questions × 3 wordings × 2 assistants (ChatGPT and Perplexity) = 54 answers. In Answersky’s Oct 7 test we asked 9 questions × 1 wording × 2 assistants × 2 repeats = 36 answers. Every answer counted here came back without an error through the assistants’ APIs with web search. The two tests asked different sets of questions and wordings so a difference between them cannot be put down to time alone. We compare sites and pages and do not claim a trend. Why we ask each question more than one way is on our how it works page.

Which cited pages name other businesses but not mine?

Open each page and mark three things. Does it name other businesses? Does it name you? What kind of page is it? The pages worth your time are the ones that name others and not you.

Some cited pages are a company describing its own offer. Whether a page like that takes outside examples depends on the page. Look for a stated way to contribute before you decide.

This step is by hand and we have done only part of it. We opened a few of the cited pages and not all of them. A site name or a web address tells you little about the page. Two business-news addresses show why. Both were cited in both tests. In Answersky’s Sep 30 test markets.businessinsider.com was cited in 2 of 54 answers. In Answersky’s Oct 7 test it was 8 of 36. In Answersky’s Sep 30 test finance.yahoo.com was cited in 1 of 54 answers. In Answersky’s Oct 7 test it was 6 of 36.

The one Yahoo Finance page we opened was a bylined advice article syndicated from another publisher. It ended with a pitch for a paid press-release service. That is not the same as a reported news story. Check who published a page and why before you treat it as a target.

What do I do for each type of site?

Act on the type of site, using what that kind of site really offers. Check its own rules and eligibility first. These are the types that showed up in our tests.

• Publisher articles and roundups (for example entrepreneur.com, forbes.com and searchenginejournal.com). Search Engine Journal says its contributors are invited by the editorial team only and that posts must be pitch-free. Whether an outlet takes expert input for an existing article depends on the page. Look for the outlet’s own stated route first. Where an outlet does take input, offer something a reader can use such as a dated data point or a short comment. In Answersky’s Sep 30 test entrepreneur.com was cited in 6 of 54 answers. In Answersky’s Oct 7 test it was 6 of 36.
• Directories and review sites. Where a site offers a free profile, claim it and fill it in the way you describe yourself on your own site, if its rules allow that. Do not pay for a listing just to get cited. In Answersky’s Sep 30 test g2.com was cited in 3 of 54 answers. In Answersky’s Oct 7 test it was 1 of 36. We could not open the g2.com page we tried. So we cannot say what it offers a business like yours.
• LinkedIn. By their address format, the cited LinkedIn pages were a mix of articles (/pulse/) and posts (/posts/). In Answersky’s Sep 30 test that was 8 articles and 4 posts. In Answersky’s Oct 7 test it was 3 articles and 2 posts. We read the address and not each page. A company page and your own posts on the topic are in your control.
• Community threads. In Answersky’s Sep 30 test reddit.com was cited in 2 of 54 answers. In Answersky’s Oct 7 test it was none of 36. Take part only where you would answer the question honestly anyway. Say who you are and follow the community’s rules.

For each page you act on, write down the date and the page. Note what you did and what happened. A site may not reply. It may not take what you offer.

How long does this take and what should I expect?

It is ongoing manual work, one page at a time. It may lead nowhere. We did not measure how long outreach takes. What gets cited changes with the question, the assistant and the day. In Answersky’s Oct 7 test 113 of the 161 sites cited had also been cited on Sep 30. In Answersky’s Oct 7 test only 119 of the 243 pages cited had also been cited on Sep 30. The same kinds of site showed up in both tests and many of the pages were not the same.

Nothing here is a promise. A listing or a mention may add a page for an assistant to find. We do not know whether it changes what an assistant says about you. The sites in our tests are the ones cited for our buyers’ questions on those two dates. They may not be the right ones for your category. Build your own list from your own questions.

Once you know where you are cited, what to fix first for AI search covers the changes to your own site.

What have we done so far?

Two things we can point to. On Oct 4 we created our LinkedIn company page and by Oct 8 our site linked to it. The other is the research in this post. That is not a full list of what we do. We are not claiming a pitch or a listing that has not happened.

If a cited page lists us in future, we will record it on our public log with its date and check later tests for any change.

To see which businesses AI named in place of yours first, try our free checker on the home page.

Free Quick Scan: 6 of your buyers’ questions on ChatGPT and Perplexity: 12 answers and who AI named instead of you.