georgia football

Direct answer

Georgia football can win AI answer engines the same way it wins the line of scrimmage: by giving the machines clean, current, machine-readable facts nobody else has bothered to publish. Depth charts in HTML. SportsTeam and SportsEvent schema. A plain-text llms.txt. Weekly refresh cycles that stay ahead of the recruiting news cycle. No trick plays. Just boring work, done on schedule.

The answer happens before the click

A fan opens ChatGPT and types: who starts at left tackle for Georgia this year. Perplexity gets the same query, worded a little differently. Neither assistant sends the fan to a search results page. It retrieves passages, ranks them, writes a sentence, and stops. Your analytics dashboard records nothing — no session, no bounce, no scroll depth. The conversation ended on someone else's screen.

That is the structural shift. Google's own guidance on creating helpful content describes systems that reward pages demonstrating first-hand expertise, and AI answer surfaces inherit a version of that logic: they want the passage that most directly resolves the question. The unit of competition is no longer the page. It's the chunk — a few hundred tokens pulled out of context and dropped into an answer.

Picture a 3,000-word game preview. Beautiful lede, embedded video, a projected depth chart sitting in paragraph fourteen behind a photo gallery. A retrieval system parsing that page has to guess which sentences matter. Now picture a bare HTML table with a single summary line above it. The table wins, and it wins by a mile, because the machine doesn't have to guess.

So the first job isn't traffic. It's being quotable to a system that will never credit you with a pageview.

Write for the passage, not the page

Here's the same content, two ways.

Version A. A gallery of eleven images. Each image is a screenshot of a position group with names overlaid in the graphic. Caption: "Projected starters." Published in August.

Version B. An HTML table with a caption, real <th> headers — Pos, No., Name, Class, Ht, Wt — and one paragraph above it: "As of October 14, Georgia's projected offensive line starters are [name] at left tackle, [name] at left guard…" Every name is also a link to that player's page.

Version B costs more discipline to maintain and looks worse in a redesign mockup. Version B is also the only one a model can read. Text in an image is invisible to a parser; text in a table with headers is a gift.

The tradeoff is honest: tables decay. Images decay too, you just never notice. When a lineman goes down in week six, Version A stays wrong for a month without anyone feeling bad about it, because nobody's editing a JPEG.

We've watched programs get cited twice as often off a single well-marked-up depth chart page than off their entire news feed — no public data measures this across athletic departments, so treat that as our working observation, not a benchmark. The mechanism is what matters: a model needs a passage where the entity, the attribute, and the value sit in one unbroken sentence. Player X plays position Y for team Z as of date D. Give it that, and it will quote you.

Give the machine an entity, not a vibe

Schema markup is where fan sites and official sites diverge. Fan sites have passion and prose. Official sites have authority and a CMS that nobody wants to touch.

The payoff for touching it is real. schema.org/SportsTeam and schema.org/SportsEvent let you state, in machine terms, that this is the team, this is the opponent, this is the date, this is the venue. schema.org/Person covers players and coaches. Google's structured data documentation explains how these graphs get consumed — and the AI surfaces draw on the same entity resolution.

The part teams skip is consistency. If your roster page says "Georgia Bulldogs," your schedule page says "UGA Football," and your news posts say "the Dawgs," you've handed the model three entities and asked it to reconcile. Give every page the same @id, the same canonical name, the same coach, the same stadium. Entity resolution is unforgiving: ambiguity produces hedging, and a hedged answer cites nobody.

This is also where agent-ready structured data earns its keep. The reason to publish a machine-readable roster isn't aesthetics — it's that an assistant answering "who's the backup quarterback" can pull a name, a class, and a number off your page instead of reconstructing it from three beat-writer tweets and a message board thread.

Freshness is the whole ballgame

Retrieval systems assemble answers from whatever they can find. When two sources disagree about your starting lineup, the model does the reasonable thing: it hedges. Reports vary. That sentence is a loss. You owned the fact and gave it away.

Content decay is the mechanism. A depth chart page published in August is accurate for two weeks and then quietly rots. A recruiting commitment page written in the spring becomes a flip story by November. Nobody on staff is tasked with noticing, because the page still gets traffic and still looks fine.

The fix is boring: assign a decay clock per page type. Depth charts and injury pages, weekly. Schedule pages, on any date change. Recruiting commitment pages, weekly through signing day. Roster pages, on any transaction. Then diff them — not by eyeballing, by comparing the rendered text to the source of truth and flagging what drifted. Content decay detection and refresh is the least glamorous capability in this entire discipline and the one that most reliably moves citation share, because you're not competing on authority. You're competing on being right this week.

One more thing. Publish the update before the aggregators do. If a fan account breaks the news and you confirm it six hours later, the model has already built its answer from the fan account. Recency signals reward whoever moved first, and you have the press release.

Measure citations, not sessions

Here's the awkward part: referral data from AI assistants is patchy. Some pass a referrer header you can see in your logs. Others send nothing at all. If you build your reporting on sessions from AI sources, you'll conclude the whole channel is a rounding error and shut down the work — which is exactly backwards.

Measure differently. Build a prompt panel: forty to sixty questions real fans actually ask, written the way they'd type them. Who's the starting quarterback. What time is kickoff. How many years does the coach have left on his contract. Who did we lose to the portal. Run that panel on a fixed schedule — weekly during the season — and record which domains get cited, which get named without a link, and which get nothing.

Over four weeks you'll see your citation rate move. You'll also see who is beating you, which is more useful than the rate itself. Maybe it's a 247Sports page with a clean table. Maybe it's a Wikipedia section. Either way you learn the shape of the gap.

A GEO and AI-visibility audit does this systematically — it maps the prompts where your program should be the canonical answer against where you're currently invisible, so you're fixing the twenty pages that matter instead of rewriting the whole site.

The 30-day runbook

Week one: check the doors. Open your robots.txt. If it disallows GPTBot or ClaudeBot, you've spent years building a museum with the front door welded shut. OpenAI documents its crawler policies at openai.com; the Model Context Protocol docs are useful background on how agents discover and call tools and content. Confirm your sitemap is clean, your canonical tags point where you think they do, and your key pages render server-side. If your depth chart only appears after JavaScript runs, some crawlers never see it.

Week two: convert assets to text. Every PDF media guide, every image-based depth chart, every "schedule" graphic. Tables, headers, one summary sentence with a date stamp. This is unglamorous and it's the highest-return week of the month.

Week three: add the graph. JSON-LD on team, event, and person pages. Same @id everywhere. Then publish an llms.txt at your root — a plain-text map telling AI agents which pages carry which canonical facts. It costs an afternoon and gives you a curated entry point instead of hoping a crawler finds the right URL.

Week four: instrument and refresh. Stand up the prompt panel. Assign decay clocks. Put a recurring calendar block on someone's name — an actual person — to update depth charts every Monday during the season.

Then repeat. Nobody wins this in a month. The programs that win it are the ones still doing week two in February.

FAQ

Does blocking GPTBot help or hurt us?

It hurts, almost always. Blocking keeps your content out of the retrieved passages that answer engines assemble their responses from, which means the model answers fan questions using someone else's page. The only case for blocking is paywalled or premium content, and even then you're trading visibility for exclusivity — a trade that's usually worse than it looks for a football program whose entire brand depends on being the first name out of the model's mouth.

How quickly do AI answers reflect a depth chart change?

There's no published refresh interval — retrieval systems rebuild indexes on their own schedules and blending recency varies by product. The practical answer is that you should assume the assistant already read yesterday's version, so the page it reads today needs to be correct today. Publishing a timestamped, unambiguous sentence is the single best lever you control.

Do we need a separate site for AI crawlers?

No. Duplicate sites split entity authority and create conflicting facts, which is the exact problem you're trying to solve. Keep one canonical source, make it crawlable, and mark it up properly. If you need a machine-friendly layer, an llms.txt file plus clean HTML on the existing domain does the job.

Can we measure how often ChatGPT cites us?

Not directly, and not completely. Referrer headers are inconsistent across assistants, so log-based reporting undercounts. The workable method is a prompt panel run on a schedule: ask the questions fans ask, log which domains appear in the answers, and track your share over time. It's sampling, not census, but the trend line is real.

Do social posts and fan forums matter for AI visibility?

They matter as corroboration, not as a primary source. When a model sees your official page and three independent fan discussions agreeing, confidence rises and the citation tends to go to the authoritative page. When they disagree, the model hedges. So the goal isn't to flood forums — it's to make sure the fact you published is the one everyone else is repeating.

Next step

You already know which pages on your site are the ones fans ask about. The question is whether an answer engine can find them, read them, and trust them today. Run a free GEO audit of your site to see exactly which prompts you own, which you're losing, and which pages need a Monday refresh.