Most brands are invisible inside AI answers for a boring reason: the model cannot confidently say what they are.
Here is the method that fixes it, in the order it should be done.
1. Get an honest baseline
Write down twenty questions a real buyer would type. Not keywords, questions, in the words they would actually use. Then run them across ChatGPT, Perplexity, Claude, and Google AI Overviews, and paste the answers verbatim into a document.
This takes an afternoon and it is the single most useful afternoon in the whole process. You will find one of three situations:
- You are not mentioned at all.
- You are mentioned with a description that is wrong.
- You are mentioned second or third, after a competitor who is worse than you.
Each situation has a different fix, and you cannot pick the fix without the transcript.
2. Fix the entity before you write anything
An entity is the machine-resolvable version of "who you are." Search engines and language models both work from entities, not strings. If three sources describe your company three different ways, a model hedges, and hedging means naming someone unambiguous instead of you.
Fixing it is unglamorous:
- Write one description. One. Forty words.
- Use it on your homepage, your about page, your LinkedIn, your Instagram bio, every directory, every podcast bio, every conference profile.
- Make your structured data say the same thing, with
sameAslinks to every profile you control. - Go find the places that describe you wrong and update them.
This is the step everyone skips because it feels like admin rather than strategy. It is the step that changes the answers most.
3. Write answer-first pages
The unit that gets quoted is a paragraph, not a page. So the format changes.
One question per URL. The question in the H1, phrased the way a person asks it. Then, immediately, an answer of forty to sixty words that makes complete sense with no surrounding context.
The test: delete everything on the page except that first paragraph. Does it still answer the title? Most pages on the internet fail that test, which is exactly why the opportunity exists.
Everything after the answer block is for humans and for corroborating detail. Write it well, but understand that its job is different.
4. Mark it up as one graph
Structured data is not a ranking factor and it is not optional either. Its job is removing ambiguity.
The mistake is scattering isolated JSON-LD blocks across pages. What works is one connected graph: your Person references your Organization, your Organization references its Services, each page references its position in the hierarchy, and every identifier stays stable.
Five types cover almost everything: Organization, Person, WebSite, Service, and FAQPage. Add BreadcrumbList and Article for structure and dates.
5. Get corroborated
This is the part that separates brands that get cited from brands that only have a nice website.
A model needs independent confirmation before it will name you. Your own site is a claim. Third-party sources are evidence. Podcasts, interviews, directory listings, industry roundups, community answers, and genuine press all count, and they count most when they describe you using the same forty words you settled on in step two.
6. Measure with a prompt set, not a rank tracker
There is no rank tracker for ChatGPT and there is not going to be one, because the output is non-deterministic. What works instead is a repeated survey.
Freeze the wording of your twenty questions. Run them monthly across each assistant. Record four things per prompt: were you mentioned, in what position, was the description accurate, and were you linked.
Monthly is right. Weekly is noise. Quarterly is too slow to connect a change to a cause.
What this actually costs
The entity work is one focused week. The first ten answer pages are two to three weeks. The structured data is a day if the entity work is already done. Corroboration is ongoing and never finishes.
Signal usually appears within four to eight weeks, which is considerably faster than competitive SEO. That is the good news, and it is temporary, the window is open because most competitors are still publishing two-thousand-word articles that bury the answer in paragraph nine.