How Perplexity, ChatGPT And Gemini Pick Their Sources

ai search optimization marketing has a long history of reporting numbers that rise while the business does not. Impressions, rankings for terms nobody buys on, traffic to pages with no commercial intent. The new channel has arrived with its own version of this, and the version is worse, because there is no independent console to check the claims against.

Observed behaviour leans toward breadth, pulling from a wider set of sources per answer than the others, and it cites forums, documentation and niche trade sources readily. It also appears comparatively responsive to freshness.

Second, prompts that presuppose a weakness: is this company expensive, are they slow, are they suitable for small clients. The answers reveal what the system believes about your reputation, and where the belief is wrong it points at a specific source you can correct.

Anything a client cannot argue with is not a report. If you cannot open the document, disagree with a conclusion and point at the evidence that contradicts it, you have been sent a reassurance rather than an analysis.

One reframing helps when presenting this internally. Report the channel as influence rather than acquisition. Acquisition framing invites a comparison against paid media on cost per lead, which this channel will lose on the reported numbers even where it is working, because most of its effect never appears as a referral. Influence framing invites the right question, which is whether more of your market arrives already knowing who you are.

Track three things over time: how often you are named, which sources get cited when you are, and which competitors appear alongside you. Movement in the second of those usually predicts movement in the first.

Run each one across the assistants your customers use, and write down the answers verbatim. Do this from a signed out session so your own history does not colour the result. What you want at the end is a simple table: which prompts named you, which named competitors, and which sources got cited.

One practical consequence of the variation between systems is worth planning for. If your customers are split across two assistants that behave differently, resist building separate programmes for each. The shared requirements account for most of the achievable outcome, and the effort spent on system specific tactics is usually better spent widening the number of third party sources that describe you correctly.

Two implications follow regardless of which system you are studying. Being findable by the underlying search step is necessary, and being worth quoting once fetched is what decides whether you are used. Almost everything actionable sits in those two requirements.

Verify the Fix Without Fooling Yourself Re-ask the same four questions quarterly rather than weekly, from a fresh signed out session. Identity work has slow feedback because scattered sources have to be re-crawled before the picture updates, and checking too often produces noise that looks like failure.

Read the Source List Before the Prose Where citations are shown, list every domain and count how often each appears. This is the single most useful output of the whole exercise, and most people skip it because the prose is more interesting.

Read the answers for confidence rather than accuracy at first. Hedged language, generic descriptions that would fit any competitor, and refusals to state a basic fact all indicate an incomplete record rather than a hostile one.

Where you do name people, make the association reciprocal. Your site names the profile, the profile links back, and ideally some independent source associates the two without either of you arranging it.

Publish the Pages Assistants Reach For Certain formats get quoted far more than others because they answer a question directly and can be lifted without distortion. Comparison pages, alternatives pages, definitional explainers, specification tables and honest pricing pages all fall into this group.

Then load your most important page with JavaScript disabled in your browser settings. If what remains is a navigation bar and no substance, that is roughly what a retrieval system reads, and it explains a great deal on its own.

Where to Get Real Language Four sources, all of which you already own. Sales call notes, where prospects describe their problem before anyone corrects their terminology. Support tickets, where customers describe things going wrong in their own words.

The prompt set is the instrument, and almost every weak measurement programme in this field has a weak prompt set at the bottom of it. Get this wrong and everything downstream measures the wrong thing with great precision.

Bring one other person from the business, ideally from sales. They will spot inaccuracies in how you are described that a marketing reader skims past, and they will tell you within minutes whether the prompts sound like real customers. That second opinion costs half an hour and prevents the most common flaw in a self run audit, which is a set of questions written in the company's own language.