How to rank on ChatGPT

ChatGPT does not have a results page, so “ranking” is the wrong frame. When it answers a question with current information, it retrieves a handful of sources, reads them, writes an answer and cites some of what it used. Your objective is to be in that handful and to be the source it quotes.

That changes what you optimise for. Classic SEO competes for a position. This competes for retrieval and extraction, and both are things you can engineer.

The two ways a model knows about you

Training. The model absorbed text about your brand during training. This is slow to influence and impossible to edit directly, and it is the reason established brands get named in answers where no browsing happened. You change it by existing in enough places, over enough time, that the pattern is learned.

Retrieval. The model searches, fetches a few pages and uses them for this specific answer. This is the tractable half. It happens live, it responds to changes in weeks rather than years, and it is where almost all practical work should go.

Attack retrieval first unless you are already a household name.

What actually gets a page retrieved and quoted

Answer the question in the first sentence. Retrieval systems chunk pages and score chunks. A chunk that opens with a direct answer scores well. Four hundred words of context before the answer means the useful chunk is buried, and the model quotes whoever got to the point.

Write self-contained claims. A sentence that depends on the previous three paragraphs cannot be lifted. A sentence that stands alone and states something specific can. “Niche edits typically cost $200 to $600 per placement” survives extraction. “As we discussed above, they are somewhat pricier” does not.

Use the question as a heading. Headings phrased as the questions people actually ask give retrieval an obvious anchor. This is why FAQ sections punch above their weight in AI citation, far more than they do in classic search.

Be specific and checkable. Numbers, ranges, dates, named constraints. Models favour concrete claims because they are easier to verify against other sources. Vague marketing prose is unquotable by construction.

Get corroborated elsewhere. This is the part most people skip, and it is the highest-leverage move on the list. A claim that appears only on your own site is a claim from an interested party. The same claim on a handful of independent sources becomes a fact the model is comfortable repeating. Corroboration is the AI-era equivalent of a backlink, and it does the same work — which is precisely why you can buy it.

Keep the machine layer clean. Schema markup, a sensible heading hierarchy, content in the HTML rather than assembled by JavaScript, and a robots policy that permits the AI crawlers you want. Blocking GPTBot and then wondering why ChatGPT never cites you is more common than it should be.

Which crawlers to allow

If you want to appear in AI answers, the relevant agents need access. GPTBot handles OpenAI’s training crawl, OAI-SearchBot handles search indexing for ChatGPT, and ChatGPT-User fetches pages during a live user session. Anthropic, Perplexity and Google run their own. Allow search indexing at minimum; allowing the training crawl as well feeds the slower half of the machine.

Check your own robots.txt before anything else. It is the cheapest possible win, and a surprising number of sites are quietly excluding the exact bots they want to reach.

Measuring it

AI answers are non-deterministic. Ask the same question five times and you may get five different citation sets. This breaks the mental model people bring from rank tracking, and it is why single checks are worthless.

What works is measuring a selection rate: take the fifty or so questions your buyers actually ask, run them repeatedly across the assistants you care about, and record how often you appear. The number that matters is the percentage, tracked over time. Movement from 3% to 12% is a real result you can put in front of a client. “ChatGPT mentioned us yesterday” is noise.

Where the real work is

Plenty of what gets sold as generative engine optimisation is classic SEO with new vocabulary, and the overlap is genuine: pages that rank get retrieved, authority still decides who is trusted, and a page nobody links to is a page nobody cites. Do that groundwork and you are already ahead of most of the market.

The distinct AI work is narrow but decisive, and it splits into three moves you can run in parallel:

• Structure for extraction. Rebuild your key pages so the answer leads, claims stand alone, and questions become headings. This is on-page and fast.
• Build corroboration across independent sources. Get the same claims about your brand placed on domains the model already trusts, so a citation stops being self-reported and starts being a fact. This is the lever most competitors never pull, and it is what our AI-citation and AI-PBN products deliver at volume.
• Feed the training layer over time. Persistent, consistent presence across many sources is what shifts the answers a model gives without browsing at all.

How to run it

• Point corroboration at the claims you want repeated. Decide the two or three statements about your brand you want ChatGPT to say back, then place those exact claims across independent sources so they line up. Consistency across sources is what makes a model comfortable repeating them.
• Vary the sources and the phrasing. A single claim echoed word-for-word across ten thin sites reads as one voice. Spread it across sources with real standing and let the wording move naturally. That is what makes the corroboration hold.
• Buy hosts with genuine authority. A placement on a domain the model already retrieves is worth ten on domains it ignores. Authority is the whole point of the placement, so pay for it.
• Pace placements and measure as you go. Track the selection rate before you start and watch it move. That is how you know the corroboration is landing and where to add more.

Start with the AI citations product to get your claims corroborated on sources the models already trust, add the AI PBN when you want a network you control feeding those claims at scale, and use the AI engine to drive the whole set. Do the on-page work yourself, buy the corroboration, and measure the selection rate climbing. That is how you get named in AI answers.