Why AI answers cite other sites about you

Whose pages decide a real AI answer?

Not the vendors'. The table at the top of this page is a question Attensira tracks for its own brand, "Peec AI alternatives", and the pages that weighed most in Perplexity's 48 answers to it in the 30 days to 4 October 2026, read over the Attensira MCP that morning. Attensira ranks the sources behind an answer by how much they decided it: their share of the citations, weighted by how high in the answer they sat.

Every one is a page comparing several products, written by someone other than Peec. Attensira was not on any of them, and our own alternatives page was not among the sources at all.

ChatGPT's answers to the same question leaned on a different set, and the source that weighed most there was peec.ai's own page for agencies, followed by alternatives roundups. Attensira was on none of those either.

This is the pattern across everything we track. In the same month, about three in four of the citations in the answers to our 61 questions went to third-party sites and about one in seven to our tracked competitors' own sites. Our own share, and how few questions it came from, is further down this page.

So the engine's view of our category is assembled almost entirely from other people's writing. That is not a quirk of our brand. An AirOps analysis of 21,311 brand mentions across GPT-5, Claude Sonnet 4.5 and Perplexity Sonar, published in October 2025, found that 85% came from external domains and 13.2% from the brand's own site.

Why do engines prefer other people's pages?

Three reasons, all about the shape of the question.

Comparison questions need comparisons. "Best X", "X alternatives" and "X vs Y" ask the engine to weigh several options. A page that already does that weighing is the most useful source it can find. A vendor's own page argues one side.

Corroboration looks like reliability. When several independent pages say the same thing about a product, an engine can state it with less risk. A claim that appears only on the vendor's site is one source with an interest.

Engines cast wider nets than search pages. Google says its AI features show a wider and more diverse set of links than a classic search, and our own answers carried about 22 links each. More slots means more room for roundups, forums and review sites alongside, or instead of, the brands themselves.

Do engines lean on the same third-party pages?

No, which doubles the work. On our "best GEO tools" question, ChatGPT's most decisive sources were xseek.io, bestgeosoftware.com, sitepoint.com, geosoftwarerankings.com and elmohq.com. Perplexity's were geoptie.com, bestgeotool.com, visby.ai, producthunt.com and seranking.com. Not one domain appeared in both lists.

Attensira's per-engine guidance draws the broader version of the same line. ChatGPT answers tend to lean on a brand's own docs and pricing pages; Claude on review sites and listicles; Perplexity, Gemini and Google AI Overviews on threads, YouTube and roundups. That guidance comes from a standard table until a workspace has enough of its own readings, at which point it switches to what the workspace's own history shows.

The consequence is practical. A generic list of "high-authority sites to get listed on" is close to useless. The list that matters is the one built from your own questions' answers, engine by engine.

What do third-party pages say when buyers ask about you by name?

Even branded answers draw on them. When Perplexity answered "what is Attensira" on 2 and 3 October 2026, its sources included our own site and docs, and also our profiles on g2.com, capterra.co.uk, pitchwall.co and linkedin.com, and a techtimes.com article.

Each of those pages carries a description of us that someone wrote at some point, and the engine treats them as corroboration. If they describe an older version of the product, the answer does too. Your branded answers are only as current as the least current profile the engines read.

That makes the first piece of third-party work unglamorous and fast: find every profile, listing and directory entry that engines cite for your branded questions, and make each one accurate.

How do you earn a place on the pages that decide answers?

Honestly, in a fixed order, starting where the odds are best.

1. Pages engines already cite that name a competitor and not you. The engine has already decided it trusts the page. The only thing missing is you. These are roundups, alternatives lists, directories and comparison articles. Attensira's off-site work always starts here and does not skip to a lower rung while one is open.

2. Pages that outrank you for the question but are not yet cited. Worth joining, with less certainty of payoff.

3. Only then, a new page. When nothing worth joining exists, publish the page that would deserve to be cited: original data, a clear comparison, a definitive answer to a question nobody answers well.

The method for each kind of page differs:

• Directory or review site

  How to get on it

  Complete the listing with accurate facts; ask real customers for reviews

  What not to do

  Invent reviews, or list features you do not have

• Roundup or alternatives post

  How to get on it

  Pitch the author with a fact the page is missing

  What not to do

  Send a template asking to be "featured"

• Forum or community thread

  How to get on it

  Answer the question genuinely, disclosing who you are

  What not to do

  Post as an undisclosed fan, or post the same reply everywhere

• Competitor's comparison page

  How to get on it

  Nothing directly; win the neutral pages around it

  What not to do

  Argue in their comments

• Open-source list or repository

  How to get on it

  Open a pull request that follows the list's rules

  What not to do

  Add yourself to lists you do not fit

Which third-party pages are not worth chasing?

Some pages look valuable and are not, at least not yet.

• Pages no engine cites for your questions. A high-authority site that never appears in your answers does nothing for them. Join it for other reasons, or later.
• Pages where you do not belong. An "alternatives to an enterprise suite" list is the wrong place for a tool for five-person teams. Getting added anyway puts a sentence about you on the page that an engine may lift into the wrong answer.
• Pages that charge for inclusion without saying so. Some lists sell their slots. If an engine cites them, being listed may help, but a paid placement that is not disclosed as one is a claim about you that you cannot stand behind.
• Your competitor's comparison page about you. You will not win a page your competitor controls. Win the neutral pages next to it.

The test for every page is the same: does an engine cite it when answering a question your buyers ask, and would a fair author include you on it?

What guardrails keep third-party work honest?

The fastest ways to lose this game are the dishonest ones. A fake review, an undisclosed forum post or an invented claim can get you named once, then get you removed from the page, and from the engine's trust, for good.

Attensira's own rules for this work are enforced in code, not left to judgement, and they are a reasonable standard for anyone:

• Nothing is sent from an Attensira domain on your behalf. Pitches go from your own mailbox.
• The first sends on any channel are yours. Until you have sent ten pitches by hand, nothing sends on its own.
• Pitches and listings have a daily cap per channel.
• A directory listing always waits for your approval, because it publishes claims under your name.
• Community answers are drafted for you to post, never posted unattended.

How do you know a placement worked?

Two checks, in order, and most teams only do the first.

The page shows you. Fetch the page and confirm your name or a link to your site is actually there. A submission is not a placement; a directory can quietly reject it.

An answer cites the page after you joined it. Until an engine has used the page in an answer since your addition, you have no evidence it helped. Attensira calls a placement verified only when both are true, and then re-reads the answers to the questions it should affect, so the result is a measured change or an honest "nothing we can prove yet".

Expect weeks. Engines have to re-read the page before your addition counts, and a single placement is rarely enough to move a question on its own. The measuring chapter covers how to read the before and after without fooling yourself.

Where should you start this week?

1. Take your five most valuable unbranded questions.
2. For each, read a few weeks of answers on each engine and list the pages they cite most.
3. Mark which of those pages name a competitor and not you.
4. Check your own profiles on every directory and review site cited for your branded questions, and correct anything out of date.
5. Pick the three pages from step 3 where you most clearly belong, and pitch them with a specific fact they are missing.

Then measure, and repeat. The next chapter, on schema and technical access, returns to your own site.